I know that I will sound whiny, but I think this is an important conversation. Also, excuse the kinda click-bait-y title.
\n\nI am new to Kaggle, having participated in just one competition before this, but I have noticed a lot of the old-timers comment how public kernels pushed them away from the site. I am one of those people who believe public kernels are great and crucial to the website's ecosystem. They have helped newcomers such as myself a lot in getting a head start and understanding the proper direction a problem can be solved in.
\n\nSo, for educational purposes, kernels are fantastic. It seems though, as with the rest of the internet, when imaginary points are involved people tend to get a bit carried away and post things just for the sake of gaining more points (an example of this is the plethora of carbon-cut blending kernels).
\n\nNormally, I wouldn't care much about such things. I understand that all these kernels that are a basic fork of each other drown out quality kernels, but I don't think that's a huge issue (people who want to read quality kernels will simply look past all the fork-of-fork-of-blending posts).
\n\nBut, people sometimes forget that this is a competition first and foremost and a lot of us put a ton of work into these, only to find ourselves drop a hundred places or more because someone decided to share a high-scoring model. I understand that, again, this is generally not a huge issue (in fact, it may be positive) since it drives us forward and pushes us to dissect the new methods and improve upon them. The problem is when people start posting great solutions right at the end of the competition. There is no time left to work and learn from these new insights, so the only thing this accomplishes is a) get the posters some shiny points, and b) help people who do nothing but fork and submit.
\n\nI am starting to understand why old-timers post about kernels pushing them away. This is not in the spirit of competition and is disappointing to see time and time again. I believe something needs to be done about this. Maybe add downvotes to kernels? This way people can voice their disappointment in a manner that will affect point-gamers. Not sure if or how this will work, but something needs to change.
\n\nAlso, a lot of people posting this type of kernels are new to the site and may not know of this \"Kaggle etiquette\". Maybe a guide should be written informing people of this?
\n\nTL;DR: Something needs to be done about people posting high-scoring models at the end of competitions.
\n\nApologies for the long post.
\n\nEDIT: Good luck gals and guys, brace for impact.
", "messages": [ { "id": 324213, "postDate": "2018-05-07T12:00:14.603Z", "content": "I know that I will sound whiny, but I think this is an important conversation. Also, excuse the kinda click-bait-y title.
\n\nI am new to Kaggle, having participated in just one competition before this, but I have noticed a lot of the old-timers comment how public kernels pushed them away from the site. I am one of those people who believe public kernels are great and crucial to the website's ecosystem. They have helped newcomers such as myself a lot in getting a head start and understanding the proper direction a problem can be solved in.
\n\nSo, for educational purposes, kernels are fantastic. It seems though, as with the rest of the internet, when imaginary points are involved people tend to get a bit carried away and post things just for the sake of gaining more points (an example of this is the plethora of carbon-cut blending kernels).
\n\nNormally, I wouldn't care much about such things. I understand that all these kernels that are a basic fork of each other drown out quality kernels, but I don't think that's a huge issue (people who want to read quality kernels will simply look past all the fork-of-fork-of-blending posts).
\n\nBut, people sometimes forget that this is a competition first and foremost and a lot of us put a ton of work into these, only to find ourselves drop a hundred places or more because someone decided to share a high-scoring model. I understand that, again, this is generally not a huge issue (in fact, it may be positive) since it drives us forward and pushes us to dissect the new methods and improve upon them. The problem is when people start posting great solutions right at the end of the competition. There is no time left to work and learn from these new insights, so the only thing this accomplishes is a) get the posters some shiny points, and b) help people who do nothing but fork and submit.
\n\nI am starting to understand why old-timers post about kernels pushing them away. This is not in the spirit of competition and is disappointing to see time and time again. I believe something needs to be done about this. Maybe add downvotes to kernels? This way people can voice their disappointment in a manner that will affect point-gamers. Not sure if or how this will work, but something needs to change.
\n\nAlso, a lot of people posting this type of kernels are new to the site and may not know of this \"Kaggle etiquette\". Maybe a guide should be written informing people of this?
\n\nTL;DR: Something needs to be done about people posting high-scoring models at the end of competitions.
\n\nApologies for the long post.
\n\nEDIT: Good luck gals and guys, brace for impact.
", "rawMarkdown": "*I know that I will sound whiny, but I think this is an important conversation. Also, excuse the kinda click-bait-y title.*\n\nI am new to Kaggle, having participated in just one competition before this, but I have noticed a lot of the old-timers comment how public kernels pushed them away from the site. I am one of those people who believe public kernels are great and crucial to the website's ecosystem. They have helped newcomers such as myself a lot in getting a head start and understanding the proper direction a problem can be solved in.\n\nSo, for educational purposes, kernels are *fantastic*. It seems though, as with the rest of the internet, when imaginary points are involved people tend to get a bit carried away and post things just for the sake of gaining more points (an example of this is the plethora of carbon-cut blending kernels).\n\nNormally, I wouldn't care much about such things. I understand that all these kernels that are a basic fork of each other drown out quality kernels, but I don't think that's a huge issue (people who want to read quality kernels will simply look past all the fork-of-fork-of-blending posts).\n\n**But**, people sometimes forget that this is a competition first and foremost and a lot of us put a ton of work into these, only to find ourselves drop a hundred places or more because someone decided to share a high-scoring model. I understand that, again, this is generally not a huge issue (in fact, it may be positive) since it drives us forward and pushes us to dissect the new methods and improve upon them. The problem is when people start posting great solutions right at the end of the competition. There is no time left to work and learn from these new insights, so the only thing this accomplishes is a) get the posters some shiny points, and b) help people who do nothing but fork and submit.\n\nI am starting to understand why old-timers post about kernels pushing them away. This is not in the spirit of competition and is disappointing to see time and time again. I believe something needs to be done about this. Maybe add downvotes to kernels? This way people can voice their disappointment in a manner that will affect point-gamers. Not sure if or how this will work, but something needs to change.\n\nAlso, a lot of people posting this type of kernels are new to the site and may not know of this \"Kaggle etiquette\". Maybe a guide should be written informing people of this?\n\n---\n\n**TL;DR**: Something needs to be done about people posting high-scoring models at the end of competitions.\n\nApologies for the long post.\n\nEDIT: Good luck gals and guys, brace for impact.", "votes": 126 }, { "id": 324898, "postDate": "2018-05-08T01:25:41.393Z", "content": "I know what we learned is much more important than the score, but many companies are using Kaggle as one of evaluation metrics for hiring data scientists. Kaggle itself has the job board, and many companies are posting jobs there.
\n\nI think Kaggle should do something to keep the reputation unless these companies are seeking \"Blending Scientists\".
", "rawMarkdown": "I know what we learned is much more important than the score, but many companies are using Kaggle as one of evaluation metrics for hiring data scientists. Kaggle itself has the job board, and many companies are posting jobs there.\n\nI think Kaggle should do something to keep the reputation unless these companies are seeking \"Blending Scientists\".", "votes": 56, "replies": [ { "id": 325254, "postDate": "2018-05-08T09:01:39.263Z", "content": "Best comment I read on this issue!
", "rawMarkdown": "Best comment I read on this issue!", "votes": 6 }, { "id": 325301, "postDate": "2018-05-08T09:36:35.117Z", "content": "I may have a bit of a simplistic mind, but the solution seems so simple to me.\n1. Only accept best single models (indeed, companies are not seeking 'Blending Scientists')\n2. Models need to be run through private kernels\n3. The score of this run is the submission (no separate csv submissions)\n4. The private kernels could possibly all be made public automatically after comp ends
", "rawMarkdown": "I may have a bit of a simplistic mind, but the solution seems so simple to me.\n1. Only accept best single models (indeed, companies are not seeking 'Blending Scientists')\n2. Models need to be run through private kernels\n3. The score of this run is the submission (no separate csv submissions)\n4. The private kernels could possibly all be made public automatically after comp ends", "votes": 5 }, { "id": 325328, "postDate": "2018-05-08T10:08:34.230Z", "content": "I think you are throwing the baby with the bath water - it would work but we would be restricted to toy models essentially - Kaggle does run certain competitions like that but I hate them as I find they are too restrictive
", "rawMarkdown": "I think you are throwing the baby with the bath water - it would work but we would be restricted to toy models essentially - Kaggle does run certain competitions like that but I hate them as I find they are too restrictive", "votes": 1 }, { "id": 325355, "postDate": "2018-05-08T10:38:23.033Z", "content": "Hmm....ok, I see. But isn't toying models exactly what companies are looking for? I think that Netflix solution also never got implemented. I prefer to learn only things that are useful in real life. But anyway, that probably just means that I have to wait for such a competition (and do kernels in the meantime) ;-).
", "rawMarkdown": "Hmm....ok, I see. But isn't toying models exactly what companies are looking for? I think that Netflix solution also never got implemented. I prefer to learn only things that are useful in real life. But anyway, that probably just means that I have to wait for such a competition (and do kernels in the meantime) ;-).", "votes": 1 } ] }, { "id": 324378, "postDate": "2018-05-07T16:38:19.373Z", "content": "Kaggle should disable kernel sharing during the last week of the competition. I think this is the most optimal solution.
", "rawMarkdown": "Kaggle should disable kernel sharing during the last week of the competition. I think this is the most optimal solution.", "votes": 58, "replies": [ { "id": 324430, "postDate": "2018-05-07T17:30:45.807Z", "content": "This honestly seems like the most straightforward solution to the problem. Downvoting kernels won't really fix the issue (though I agree it's a nice to have feature). And coming up with a filtering mechanism is a game of cat-and-mouse, since noise can be added to the output. Better just to shut it down once at the same time that team merge deadline occurs. Alternatively, use something like Kaggle Kernel-Ranking, where users have to hit Kernel-Expert, etc. before they can open new Kernels in the last week.
", "rawMarkdown": "This honestly seems like the most straightforward solution to the problem. Downvoting kernels won't really fix the issue (though I agree it's a nice to have feature). And coming up with a filtering mechanism is a game of cat-and-mouse, since noise can be added to the output. Better just to shut it down once at the same time that team merge deadline occurs. Alternatively, use something like Kaggle Kernel-Ranking, where users have to hit Kernel-Expert, etc. before they can open new Kernels in the last week.", "votes": 6 }, { "id": 324470, "postDate": "2018-05-07T18:18:58.317Z", "content": "+1\nI would also vote for disabling public kernels in last competition week. Of course that probably would not prevent someone hunting for cheap up-votes from e.g. posting links to their Githubs in discussions, but I think that would be a step into the right direction. And discussions at least can be downvoted, if high scoring solution is posted in the last day of the competition ;)
", "rawMarkdown": "+1\nI would also vote for disabling public kernels in last competition week. Of course that probably would not prevent someone hunting for cheap up-votes from e.g. posting links to their Githubs in discussions, but I think that would be a step into the right direction. And discussions at least can be downvoted, if high scoring solution is posted in the last day of the competition ;)", "votes": 8 }, { "id": 324473, "postDate": "2018-05-07T18:20:52.810Z", "content": "My concern about blocking kernels the last week would be the same as yours - Github repos. I think posting a high scoring solution on that would be even more unfair because its not as transparent and clear as the Kaggle kernel ecosystem.
", "rawMarkdown": "My concern about blocking kernels the last week would be the same as yours - Github repos. I think posting a high scoring solution on that would be even more unfair because its not as transparent and clear as the Kaggle kernel ecosystem.", "votes": 3 }, { "id": 324478, "postDate": "2018-05-07T18:25:52.907Z", "content": "I'm not worried about github repos. For one, that would be private sharing so they (should) end up disqualified. I recall in a past competition, some people were having a discussion on KaggleNoobs slack and it started getting to code sharing were told they had bring it onto the forums---which they did---so I don't imagine github being much different.
\n\nMoreover, one has to look at the intent behind these perpetrators. Their goal is to either get a lot of upvotes, or to destroy the integrity of the competition. In the former case, disabling kernels w/ team merger deadline solves it. In the later case, it makes it a LOT harder to broadcast results to the masses, so net-net win imo.
", "rawMarkdown": "I'm not worried about github repos. For one, that would be private sharing so they (should) end up disqualified. I recall in a past competition, some people were having a discussion on KaggleNoobs slack and it started getting to code sharing were told they had bring it onto the forums---which they did---so I don't imagine github being much different.\n\nMoreover, one has to look at the intent behind these perpetrators. Their goal is to either get a lot of upvotes, or to destroy the integrity of the competition. In the former case, disabling kernels w/ team merger deadline solves it. In the later case, it makes it a LOT harder to broadcast results to the masses, so net-net win imo.", "votes": 3 }, { "id": 324479, "postDate": "2018-05-07T18:26:22.207Z", "content": "Just disable links in the last week in the discussion pages too - at least you can flag discussion posts
", "rawMarkdown": "Just disable links in the last week in the discussion pages too - at least you can flag discussion posts", "votes": 2 }, { "id": 324606, "postDate": "2018-05-07T21:08:40.027Z", "content": "@alijs But these would be clear rule violations and threat of disqualification should prevent them
", "rawMarkdown": "@alijs But these would be clear rule violations and threat of disqualification should prevent them", "votes": 1 }, { "id": 324623, "postDate": "2018-05-07T21:30:37.943Z", "content": "This is clearly the best solution. It allows for shared context, but not shared solutions.
", "rawMarkdown": "This is clearly the best solution. It allows for shared context, but not shared solutions." }, { "id": 1600274, "postDate": "2021-11-30T08:48:55.357Z", "content": "This aged well
", "rawMarkdown": "This aged well" } ] }, { "id": 325437, "postDate": "2018-05-08T12:16:21.380Z", "content": "Anyone help me debug this? It looks quiet runnable for kaggleš
\n\nimport numpy as np\nimport warnings\nfrom pandas.tseries.offsets import DateOffset\n\ndef win_medals():\n '''\n Only works to silver and bronze, have fun!\n '''\n if Competition == 'kaggle':\n assert Remaning_hours_to_deadline <= DateOffset(hours=12), \"Too early man,it's not time!\"\n\n grab_a_coffee()\n click_kernels_button()\n csvs = download_highest_score_csv(n=10)\n submits = do_blending(csvs, random_state=np.random.randint(1,100))\n submit_predictions(submits)\n\n if good_luck:\n return SILVER\n else:\n return BRONZE\n else:\n pass\n\nif __name__ == 'main':\n medals = win_medals()\n if entire_running_time >= DateOffset(hours=4)\n warnings.warn('You are wasting too much time! You should be quicker, try next time use only brute force without brain!')\nHAHA!
", "rawMarkdown": "HAHA!", "votes": 5 }, { "id": 325532, "postDate": "2018-05-08T14:08:16.997Z", "content": ":) good one @Wenjie
", "rawMarkdown": ":) good one @Wenjie", "votes": 1 }, { "id": 325535, "postDate": "2018-05-08T14:10:03.393Z", "content": "Why are you so xiu~
", "rawMarkdown": "Why are you so xiu~", "votes": 1 }, { "id": 325537, "postDate": "2018-05-08T14:14:16.517Z", "content": "can I use this code at work or only at kaggle ?
", "rawMarkdown": "can I use this code at work or only at kaggle ? ", "votes": 2 }, { "id": 325560, "postDate": "2018-05-08T14:48:12.913Z", "content": "This was awesome, can't stop laughing! The best thing that happened after the yesterday's deadline :) Thanks...
", "rawMarkdown": "This was awesome, can't stop laughing! The best thing that happened after the yesterday's deadline :) Thanks...", "votes": 2 }, { "id": 325587, "postDate": "2018-05-08T15:45:59.933Z", "content": "Hahaha~ Thank you guys for all these comments!~ Life is full of hopes and joys, right? I believe we who really think independently and work hard earn most whether for short term or long term. Hope to meet you guys in the following competition or maybe form a team to enjoy the learning process/competition together!š
", "rawMarkdown": "Hahaha~ Thank you guys for all these comments!~ Life is full of hopes and joys, right? I believe we who really think independently and work hard earn most whether for short term or long term. Hope to meet you guys in the following competition or maybe form a team to enjoy the learning process/competition together!š", "votes": 7 }, { "id": 325848, "postDate": "2018-05-09T00:46:32.207Z", "content": "Cool! I want to use this for next competition!, but I hope Kaggle conducts competitions appropriately in the future.
", "rawMarkdown": "Cool! I want to use this for next competition!, but I hope Kaggle conducts competitions appropriately in the future.", "votes": 2 }, { "id": 325864, "postDate": "2018-05-09T01:27:59.230Z", "content": "So pi~~~
", "rawMarkdown": "So pi~~~", "votes": 1 }, { "id": 326100, "postDate": "2018-05-09T09:27:46.873Z", "content": "Great code ! Do you have the same in R ? ;)
", "rawMarkdown": "Great code ! Do you have the same in R ? ;)", "votes": 2 }, { "id": 326186, "postDate": "2018-05-09T12:29:53.887Z", "content": "Cant find the submit button, please help!
", "rawMarkdown": "Cant find the submit button, please help!", "votes": 1 }, { "id": 326220, "postDate": "2018-05-09T13:08:55.150Z", "content": "ęę¶ēēē®
", "rawMarkdown": "ęę¶ēēē®" }, { "id": 326252, "postDate": "2018-05-09T13:38:52.087Z", "content": "You're definitely not äŗŗé é©, You're ēēē®.
", "rawMarkdown": "You're definitely not äŗŗé é©, You're ēēē®." } ] }, { "id": 324664, "postDate": "2018-05-07T22:03:30.553Z", "content": "Many people have suggested to block kernels in the last week. We have been saying this for ages. Why kaggle hasn't acted on this?
\n\nI am curious on the reasoning that kernels are still allowed in the last week ... I have honestly not seen much opposition for this suggestion (if any).
\n\nOn the contrary I think kaggle seems to want this. Kernels and discussions are gamified and you can get points/ranks from these elements . If someone is not very high in competitions or he/she targets to get master/grandmaster through kernels, he/she can unleash a high scoring kernel near the end to get points/votes.
", "rawMarkdown": "Many people have suggested to block kernels in the last week. We have been saying this for ages. Why kaggle hasn't acted on this?\n\n I am curious on the reasoning that kernels are still allowed in the last week ... I have honestly not seen much opposition for this suggestion (if any). \n\nOn the contrary I think kaggle seems to want this. Kernels and discussions are gamified and you can get points/ranks from these elements . If someone is not very high in competitions or he/she targets to get master/grandmaster through kernels, he/she can unleash a high scoring kernel near the end to get points/votes.", "votes": 42, "replies": [ { "id": 324683, "postDate": "2018-05-07T22:17:47.073Z", "content": "I believe highly ranked kernel publication at the last minute gives to the game more stamina. It is like throwing some oil on the fire. It fires up more and makes the game more popular... It reminds me ebay auction where people wait until the last minute to make their bid...
", "rawMarkdown": "I believe highly ranked kernel publication at the last minute gives to the game more stamina. It is like throwing some oil on the fire. It fires up more and makes the game more popular... It reminds me ebay auction where people wait until the last minute to make their bid...", "votes": -11 }, { "id": 324685, "postDate": "2018-05-07T22:18:26.917Z", "content": "The only reason I can think of is that it inflates the scores which looks better for kaggle. I have wracked my brains for a less cynical explanation but I cannot find one
", "rawMarkdown": "The only reason I can think of is that it inflates the scores which looks better for kaggle. I have wracked my brains for a less cynical explanation but I cannot find one", "votes": 2 }, { "id": 324689, "postDate": "2018-05-07T22:22:05.457Z", "content": "\n\n\neric wrote
\n \n\n\nI believe highly ranked kernel publication at the last minute gives to the game more stamina. It is like throwing some oil on the fire. It fires up more and makes the game more popular... It reminds me ebay auction where people wait until the last minute to make their bid...
\n
I don't think that several month long data science competitions that involve 200 million datapoints should be reminiscent of last minute ebay auctions.
", "rawMarkdown": "\n> **eric wrote**\n> \n> > I believe highly ranked kernel publication at the last minute gives to the game more stamina. It is like throwing some oil on the fire. It fires up more and makes the game more popular... It reminds me ebay auction where people wait until the last minute to make their bid...\n\nI don't think that several month long data science competitions that involve 200 million datapoints should be reminiscent of last minute ebay auctions.", "votes": 16 }, { "id": 324692, "postDate": "2018-05-07T22:23:55.640Z", "content": "Hope that this kernel is an overfitted joke and it will be the epic competition.
", "rawMarkdown": "Hope that this kernel is an overfitted joke and it will be the epic competition.", "votes": 4 }, { "id": 324723, "postDate": "2018-05-07T22:55:32.507Z", "content": "Scirpus, agree that it inflates scores and makes the kaggle community look smarter!
", "rawMarkdown": "Scirpus, agree that it inflates scores and makes the kaggle community look smarter!", "votes": -6 }, { "id": 325132, "postDate": "2018-05-08T06:44:51.960Z", "content": "Kernels are one part of the problem. What about the discussions? If I remember correctly, a huge spoiler in the Instacart Market Basket competition was shared via github. In the end I finished some ~ 150 places lower, and lost most of the motivation after more than a month of active participation.
", "rawMarkdown": "Kernels are one part of the problem. What about the discussions? If I remember correctly, a huge spoiler in the Instacart Market Basket competition was shared via github. In the end I finished some ~ 150 places lower, and lost most of the motivation after more than a month of active participation.", "votes": 1 }, { "id": 325316, "postDate": "2018-05-08T09:49:26.193Z", "content": "As @inversion mentioned in this thread this is a very serious issue that we are taking very seriously within Kaggle and have been thinking about for a long time, obviously.
\n\nJust to add a little bit of context here. For about a year we used to disable kernel submission to competitions during the last week of a competition. Unfortunately this didn't stop people sharing the code on kernels anyway (or in discussion posts) and it caused a lot of users to write to us upset because they didn't understand why submission wasn't working.
\n\nIn June last year we added private kernels to the site as the default, this has opened up some more possibilities that we are going to be actively talking about. But its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.
", "rawMarkdown": "As @inversion mentioned in this thread this is a very serious issue that we are taking very seriously within Kaggle and have been thinking about for a long time, obviously.\n\nJust to add a little bit of context here. For about a year we used to disable kernel submission to competitions during the last week of a competition. Unfortunately this didn't stop people sharing the code on kernels anyway (or in discussion posts) and it caused a lot of users to write to us upset because they didn't understand why submission wasn't working.\n\nIn June last year we added private kernels to the site as the default, this has opened up some more possibilities that we are going to be actively talking about. But its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.", "votes": 11 }, { "id": 325333, "postDate": "2018-05-08T10:12:22.783Z", "content": "@ myles. Thanks for the writing in this forum.
\n\nI have a few specific questions.
\n\nWhat has happened in literally the last few hours is that hundreds of users have uploaded the same file as their input data file. Does Kaggle consider this as a violation of rules? And by violation I mean something that is serious enough for removal from the competition.
\n\nIf a kernel was public and everyone ran that same kernel for a similar score there is nothing anyone can do about it. But here, people have directly uploaded the file as their own private data set. Isn't the nature of this problem a bit more grave and shouldn't this be penalized?
\n\nPlease do let us Kagglers know where Kaggle stands on this because if this is not a violation perhaps in the next competition when someone leaks a kernel I'll be sure to blend it myself - especially since I wouldn't run the risk of disqualification.
\n\nThe thought of safeguarding my position did occur to me yesterday but I DID NOT USE the file only, only because I was SO SURE that Kaggle would consider this a violation.
\n\nRegards\nShanth
", "rawMarkdown": "@ myles. Thanks for the writing in this forum. \n\nI have a few specific questions. \n\nWhat has happened in literally the last few hours is that hundreds of users have uploaded the same file as their input data file. Does Kaggle consider this as a violation of rules? And by violation I mean something that is serious enough for removal from the competition. \n\nIf a kernel was public and everyone ran that same kernel for a similar score there is nothing anyone can do about it. But here, people have directly uploaded the file as their own private data set. ***Isn't the nature of this problem a bit more grave and shouldn't this be penalized?*** \n\nPlease do let us Kagglers know where Kaggle stands on this because if this is not a violation perhaps in the next competition when someone leaks a kernel I'll be sure to blend it myself - especially since I wouldn't run the risk of disqualification. \n\nThe thought of safeguarding my position did occur to me yesterday but I DID NOT USE the file only, only because I was SO SURE that Kaggle would consider this a violation. \n\nRegards\nShanth ", "votes": 2 }, { "id": 325394, "postDate": "2018-05-08T11:22:00.467Z", "content": "hey @Myles O'Neill
\n\nThank you for the response - I am glad kaggle is looking into this. I agree that if someone is determined to pass on the information , he/she will find a way to do it - but it is a whole different story when you essentially encourage it ( with potential kernel upvotes).
\n\nIt is a culture thing too ... People should somehow be aware that this is not how things should work , even if it is not strictly on the rules. I dont think the culture is there now.
", "rawMarkdown": "hey @Myles O'Neill\n\nThank you for the response - I am glad kaggle is looking into this. I agree that if someone is determined to pass on the information , he/she will find a way to do it - but it is a whole different story when you essentially encourage it ( with potential kernel upvotes). \n\nIt is a culture thing too ... People should somehow be aware that this is not how things should work , even if it is not strictly on the rules. I dont think the culture is there now. \n", "votes": 8 }, { "id": 325400, "postDate": "2018-05-08T11:33:16.490Z", "content": "\n\n\nMyles O'Neill wrote
\n \n\n\n \nAs @inversion mentioned in this thread this is a very serious issue that we are taking very seriously within Kaggle and have been thinking about for a long time, obviously.
\nJust to add a little bit of context here. For about a year we used to disable kernel submission to competitions during the last week of a competition. Unfortunately this didn't stop people sharing the code on kernels anyway (or in discussion posts) and it caused a lot of users to write to us upset because they didn't understand why submission wasn't working.
\n \nIn June last year we added private kernels to the site as the default, this has opened up some more possibilities that we are going to be actively talking about. But its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.
\n
It exists solution that may works. If some people are not mature enough to understand what a 'fair competition' is then it have to be imposed. Like in that case - disabling kernels is not enough because of discussion sharing possibility. Thus, I think, it should be forbidden to share your kernels / solutions / csv/ etc. in any way during last week of competition. It is very strict, but efficient.
\n\nI wish that everyone can understand 'fair play' but this is the ideal world - in which we are not living :(
\n\nThanks for your involvement in that case!
", "rawMarkdown": "\n> **Myles O'Neill wrote**\n> \n> > As @inversion mentioned in this thread this is a very serious issue that we are taking very seriously within Kaggle and have been thinking about for a long time, obviously.\n> \n> Just to add a little bit of context here. For about a year we used to disable kernel submission to competitions during the last week of a competition. Unfortunately this didn't stop people sharing the code on kernels anyway (or in discussion posts) and it caused a lot of users to write to us upset because they didn't understand why submission wasn't working.\n> \n> In June last year we added private kernels to the site as the default, this has opened up some more possibilities that we are going to be actively talking about. But its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.\n\nIt exists solution that may works. If some people are not mature enough to understand what a 'fair competition' is then it have to be imposed. Like in that case - disabling kernels is not enough because of discussion sharing possibility. Thus, I think, it should be forbidden to share your kernels / solutions / csv/ etc. in any way during last week of competition. It is very strict, but efficient.\n\nI wish that everyone can understand 'fair play' but this is the ideal world - in which we are not living :(\n\nThanks for your involvement in that case! " }, { "id": 325404, "postDate": "2018-05-08T11:39:01.923Z", "content": "Shutdown Discussion in the last week too - draconian - probably - but it just might work
", "rawMarkdown": "Shutdown Discussion in the last week too - draconian - probably - but it just might work", "votes": 2 }, { "id": 325423, "postDate": "2018-05-08T11:59:55.383Z", "content": "I would like to avoid this method, but you have a point ;)
", "rawMarkdown": "I would like to avoid this method, but you have a point ;)", "votes": 1 }, { "id": 325482, "postDate": "2018-05-08T13:06:57.187Z", "content": "\n\n\nMyles O'Neill wrote
\n \nBut its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.
\n
I believe there a huge gap between users checking kernels and users reading posts on discussion forum. So blocking kernels can certainly reduce the major impact. I also think it's a good idea let the community decide what to do with such kernels or links posted on last day if Kaggle can't take any solid action on LB destroyers.
\n\nFor example, allowing specific user group to delete the kernel or link posted on last day . This user group can be something like participants with contributor or above badge and having atleast 50 submissions.
\n\nLet's say if 20 users from this special user group indicates that kernel/link is harmful to competition, they can simply take it down.
", "rawMarkdown": "\n> **Myles O'Neill wrote**\n> \n> But its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.\n\nI believe there a huge gap between users checking kernels and users reading posts on discussion forum. So blocking kernels can certainly reduce the major impact. I also think it's a good idea let the community decide what to do with such kernels or links posted on last day if Kaggle can't take any solid action on LB destroyers. \n\nFor example, allowing specific user group to delete the kernel or link posted on last day . This user group can be something like participants with contributor or above badge and having atleast 50 submissions. \n\nLet's say if 20 users from this special user group indicates that kernel/link is harmful to competition, they can simply take it down.", "votes": 2 }, { "id": 325915, "postDate": "2018-05-09T04:02:16.533Z", "content": "I think a combination between disabling kernels in the last week and banning users who post/share kernels in discussions in the last week should solve the issue. If people try to get around the rules/system then they should be banned, similar to users who create multiple accounts to get extra submissions. Only issue that would leave is people with public githubs... but at that point I feel like it would be more that they were just careless on keeping their github private than maliciously sharing code.
", "rawMarkdown": "I think a combination between disabling kernels in the last week and banning users who post/share kernels in discussions in the last week should solve the issue. If people try to get around the rules/system then they should be banned, similar to users who create multiple accounts to get extra submissions. Only issue that would leave is people with public githubs... but at that point I feel like it would be more that they were just careless on keeping their github private than maliciously sharing code." }, { "id": 326388, "postDate": "2018-05-09T16:16:50.380Z", "content": "Check this out, the leaderboard progression animation: https://www.kaggle.com/inversion/talkingdata-leaderboard-progression/code
", "rawMarkdown": "Check this out, the leaderboard progression animation: https://www.kaggle.com/inversion/talkingdata-leaderboard-progression/code" } ] }, { "id": 324476, "postDate": "2018-05-07T18:24:02.497Z", "content": "This is a discussion point that gets re-ignited every few months, and becomes more challenging as Kaggle continues to add more functionality to Kernels and Datasets.
\n\nWe are following this closely, and will be having focused discussions internally on how we can both (a) continue to provide value-added tools to the data science community as well as (b) minimize the amount of frustration that might arise during competitions because of these tools.
\n\nPlease continue to post your thoughts and suggests in this thread. Your frank feedback is very much appreciated.
", "rawMarkdown": "This is a discussion point that gets re-ignited every few months, and becomes more challenging as Kaggle continues to add more functionality to Kernels and Datasets.\n\nWe are following this closely, and will be having focused discussions internally on how we can both (a) continue to provide value-added tools to the data science community as well as (b) minimize the amount of frustration that might arise during competitions because of these tools.\n\nPlease continue to post your thoughts and suggests in this thread. Your frank feedback is very much appreciated.", "votes": 37, "replies": [ { "id": 324504, "postDate": "2018-05-07T18:51:25.820Z", "content": "@ Inversion
\n\nWe exchanged a couple of messages a few days ago where you gave me confirmation that CSV files can be used for submission directly as long as they are NOT shared outside the team and are maintained as PRIVATE data sets. \n CSV CANNOT BE SHARED
\n\nUnfortunately, someone chose to IGNORE that rule. In fact, I shared this above link with the person who had posted the kernel and within a few minutes the kernel was taken down.
\n\nCan the Kaggle Team do something about this Please?
\n\nIf someone published a kernel and everyone ran it to get to this score, I guess there is nothing Kaggle can do about it. But , in this case the CSV file was simply downloaded and then uploaded again to get to an output.
\n\nThe csv that produces the 0.9811 score with the EXACT SAME OUTPUT would have been by multiple people. Is this not a violation of competition rules? Again, running a public kernel for a high score is not in the spirit of the competition either but what has happened this time seems to be like an outright violation.
\n\nWould it be fair to disqualify anyone having the 0.9811 file in their Data files? Will the Kaggle team be able to do that in the evaluation process? If this process may take time, those of us who worked hard for more than a month will be more than happy to wait for a few more days till the final ranks come out.
\n\nI very sincerely request the Kaggle team to take this into consideration while evaluating the final competition results.
\n\nI have learnt much from Kaggle and will continue to do so. Me and a lot of others will hope for a favourable resolution on this issue. I will accept any final decision that Kaggle makes.
\n\nRegards\nShanth
", "rawMarkdown": "@ Inversion \n\nWe exchanged a couple of messages a few days ago where you gave me confirmation that CSV files can be used for submission directly as long as they are NOT shared outside the team and are maintained as PRIVATE data sets. \n [CSV CANNOT BE SHARED][1]\n\nUnfortunately, someone chose to IGNORE that rule. In fact, I shared this above link with the person who had posted the kernel and within a few minutes the kernel was taken down.\n\nCan the Kaggle Team do something about this Please?\n\nIf someone published a kernel and everyone ran it to get to this score, I guess there is nothing Kaggle can do about it. But , in this case the CSV file was simply downloaded and then uploaded again to get to an output. \n\nThe csv that produces the 0.9811 score with the EXACT SAME OUTPUT would have been by multiple people. Is this not a violation of competition rules? Again, running a public kernel for a high score is not in the spirit of the competition either but what has happened this time seems to be like an outright violation.\n\n***Would it be fair to disqualify anyone having the 0.9811 file in their Data files?*** Will the Kaggle team be able to do that in the evaluation process? If this process may take time, those of us who worked hard for more than a month will be more than happy to wait for a few more days till the final ranks come out. \n\n I very sincerely request the Kaggle team to take this into consideration while evaluating the final competition results. \n\nI have learnt much from Kaggle and will continue to do so. Me and a lot of others will hope for a favourable resolution on this issue. I will accept any final decision that Kaggle makes.\n\nRegards\nShanth\n [1]: https://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/51142#324464", "votes": 6 }, { "id": 324505, "postDate": "2018-05-07T18:52:56.783Z", "content": "@inversion Thank you for your kindness. I suggest you delete copy & paste, download & submit from leaderboard. It is simple to detect, solve the most frustrating part which is people who does nothing get better ranks. Same repeated result on leaderboard doesnāt benefit anyone except lazy cheaters.
", "rawMarkdown": "@inversion Thank you for your kindness. I suggest you delete copy & paste, download & submit from leaderboard. It is simple to detect, solve the most frustrating part which is people who does nothing get better ranks. Same repeated result on leaderboard doesnāt benefit anyone except lazy cheaters.", "votes": 4 }, { "id": 324509, "postDate": "2018-05-07T18:57:20.717Z", "content": "Those suggestions are good @shanth84
\n\nBut rather than disqualifying them, just disqualify that entry alone. A lot of hard workers who got to 0.9800 or wherever they got to succumbed to the infamous \"9811\" kernel and submitted that too, along with their own predictions out of frustration. I don't think those ppl should be DQ'd. Just that submission.
", "rawMarkdown": "Those suggestions are good @shanth84\n\nBut rather than disqualifying them, just disqualify that entry alone. A lot of hard workers who got to 0.9800 or wherever they got to succumbed to the infamous \"9811\" kernel and submitted that too, along with their own predictions out of frustration. I don't think those ppl should be DQ'd. Just that submission.", "votes": 6 }, { "id": 324513, "postDate": "2018-05-07T19:02:02.360Z", "content": "I think that feature engineering is needed ;) ... or just a simple feature to be honest. As one of kagglers mentioned (I am deeply sorry I've just remember your keynote not a nickname :( ) withdrawal of uploading kernels ~1 week before deadline will solve the case.
", "rawMarkdown": "I think that feature engineering is needed ;) ... or just a simple feature to be honest. As one of kagglers mentioned (I am deeply sorry I've just remember your keynote not a nickname :( ) withdrawal of uploading kernels ~1 week before deadline will solve the case.", "votes": 2 }, { "id": 324514, "postDate": "2018-05-07T19:02:09.223Z", "content": "@authman Disqualifying only that entry makes sense.. But already there are blends of blends with a much higher score!!
", "rawMarkdown": "@authman Disqualifying only that entry makes sense.. But already there are blends of blends with a much higher score!!", "votes": 3 }, { "id": 324515, "postDate": "2018-05-07T19:02:58.637Z", "content": "+1
", "rawMarkdown": "+1" }, { "id": 324517, "postDate": "2018-05-07T19:05:42.230Z", "content": "@ authman
\n\nI will leave it to Kaggle to decide what they want to do about this.
\n\nI don't have a medal yet. I got from 0.9797 to 0.9800 in my very last submission today even as all this was unfolding. Yet I chose NOT TO USE the 0.9811 file.
\n\nI am only questioning the idea of using a fully developed output file to improve one's score.
\n\n:)
\n\nCheers\nShanth
", "rawMarkdown": "@ authman \n\nI will leave it to Kaggle to decide what they want to do about this. \n\nI don't have a medal yet. I got from 0.9797 to 0.9800 in my very last submission today even as all this was unfolding. Yet I chose NOT TO USE the 0.9811 file.\n\nI am only questioning the idea of using a fully developed output file to improve one's score. \n\n:) \n\nCheers\nShanth \n", "votes": 4 }, { "id": 324518, "postDate": "2018-05-07T19:07:04.450Z", "content": "Hi @Shanth -
\n\nI see where there was confusion in my response. I was responding directly to your question about having a private kernel. If your kernel is private, you cannot share it, or the output of it, with anyone that is not on your team.
\n\nPublicly-shared kernels, and the output to such, are available to anyone. This creates the current difficulty.
", "rawMarkdown": "Hi @Shanth -\n\nI see where there was confusion in my response. I was responding directly to your question about having a _private_ kernel. If your kernel is private, you cannot share it, or the output of it, with anyone that is not on your team. \n\nPublicly-shared kernels, and the output to such, are available to anyone. This creates the current difficulty. ", "votes": 1 }, { "id": 324521, "postDate": "2018-05-07T19:11:03.833Z", "content": "@inversion In Data Science Bowl 2018, admins could delete people from leaderboard if āthey are against the spirit of the competitionā. Can kaggle do the same here? I think now itās the time.
", "rawMarkdown": "@inversion In Data Science Bowl 2018, admins could delete people from leaderboard if āthey are against the spirit of the competitionā. Can kaggle do the same here? I think now itās the time.", "votes": 6 }, { "id": 324537, "postDate": "2018-05-07T19:24:17.687Z", "content": "I would agree with @authman.
\n\nI think disable submissions (except who originally posted it on kernels) that are completely identical to the publicly available csv seems to be a better idea. Otherwise, if posting csv is considered illegal sharing, then a lot of hard working kernel contributors will be disqualified, since many of the kernels have posted the results (just not as high as 0.9811 or as late in the competition).
\n\nAlthough the situation might be very frustrating, we should be very cautious about disqualifying kagglers from competition, especially if the rule is not well stated, emphasized and enforced before. I think any rule change or re-clarification and enforcement should be done for future competition with a clear statement at the beginning of the competition. (But this is just my opinion and up for discussion).
\n\nOtherwise, Kaggle can wipe out half of the lead board with a snap of his fingerā¦
", "rawMarkdown": "I would agree with @authman. \n\nI think disable submissions (except who originally posted it on kernels) that are completely identical to the publicly available csv seems to be a better idea. Otherwise, if posting csv is considered illegal sharing, then a lot of hard working kernel contributors will be disqualified, since many of the kernels have posted the results (just not as high as 0.9811 or as late in the competition).\n\nAlthough the situation might be very frustrating, we should be very cautious about disqualifying kagglers from competition, especially if the rule is not well stated, emphasized and enforced before. I think any rule change or re-clarification and enforcement should be done for future competition with a clear statement at the beginning of the competition. (But this is just my opinion and up for discussion). \n\nOtherwise, Kaggle can wipe out half of the lead board with a snap of his fingerā¦\n", "votes": 4 }, { "id": 324544, "postDate": "2018-05-07T19:28:52.580Z", "content": "@ inversion
\n\nThanks for the response. Hope you can speak with the rest of the Kaggle team to consider possible options.
\n\nYour response is completely understandable if someone just ran the kernel and generated their own outputs - I guess there would be no way to stop or monitor it.
\n\nBut having a public kernels output AS A DATA SOURCE and then submitting it just doesn't seem right. It is clearly against the spirit of the competition and the platform that Kaggle is.
\n\nHope the Kaggle team can do something about it .
\n\nRegards\nPrasanth
", "rawMarkdown": "@ inversion \n\nThanks for the response. Hope you can speak with the rest of the Kaggle team to consider possible options.\n\n Your response is completely understandable if someone just ran the kernel and generated their own outputs - I guess there would be no way to stop or monitor it. \n\nBut having a public kernels output AS A DATA SOURCE and then submitting it just doesn't seem right. It is clearly against the spirit of the competition and the platform that Kaggle is. \n\nHope the Kaggle team can do something about it .\n\nRegards\nPrasanth ", "votes": 1 }, { "id": 324553, "postDate": "2018-05-07T19:34:40.670Z", "content": "@ RLstat - I Like the Thanos reference :)
", "rawMarkdown": "@ RLstat - I Like the Thanos reference :)", "votes": 1 }, { "id": 324607, "postDate": "2018-05-07T21:09:01.270Z", "content": "@Inversion\nEarly on kernels help newer Kagglers get up to speed. Right before submission, they provide an unfair advantage.
\n\nI strongly recommend disabling or delaying kernel submissions during the last week. Competitors should be given a chance to think about the problem on their own.
", "rawMarkdown": "@Inversion\nEarly on kernels help newer Kagglers get up to speed. Right before submission, they provide an unfair advantage. \n\nI strongly recommend disabling or delaying kernel submissions during the last week. Competitors should be given a chance to think about the problem on their own.", "votes": 1 }, { "id": 324608, "postDate": "2018-05-07T21:09:05.783Z", "content": "I am thinking do we really need kernel for these competitions. For people who are really working on the problem, discussion is enough with no need showing all code. For new comers like me, kernels in those competitions with no money involved should already be enough.
", "rawMarkdown": "I am thinking do we really need kernel for these competitions. For people who are really working on the problem, discussion is enough with no need showing all code. For new comers like me, kernels in those competitions with no money involved should already be enough.", "votes": 2 }, { "id": 325064, "postDate": "2018-05-08T04:43:08.823Z", "content": "Inversion : I gave it a thought and here are my 2 cents. \"Like there is a team merger deadline , let's have a kernel submission deadline as well (Maybe same as team merger) \" That way everyone gets to learn as well and if a competition lasts for 2 months , i think 6 weeks are enough to generate ideas in kernels. That way this level of frustration can be avoided.
", "rawMarkdown": "Inversion : I gave it a thought and here are my 2 cents. \"Like there is a team merger deadline , let's have a kernel submission deadline as well (Maybe same as team merger) \" That way everyone gets to learn as well and if a competition lasts for 2 months , i think 6 weeks are enough to generate ideas in kernels. That way this level of frustration can be avoided." }, { "id": 325101, "postDate": "2018-05-08T05:55:04.120Z", "content": "A couple of rules that'd make sense to me:
\n\n\n\n\nPlease continue to post your thoughts and suggests in this thread. Your frank feedback is very much appreciated.
\n
One small suggestion:
\n\nWith data this size, tied scores are nearly certain to be identical CSVs.
\n\nBreaking ties by time of submission becomes less meaningful/appropriate the more people submit a publicly shared CSV.
\n\nWhy not group identical entries into one implicit team rank?
\n\nIf you keep the original number of teams for medals calculations (as in two stage competitions) that would move some teams upwards into the medals zones. (Actually it may reward some admirable people - those who worked hard and gained an honest score just a hair below the shared solution but still did not use it.)
\n\nPerhaps even dilute the points of the new merged teams in the same way actual teams are penalized by size. (More frankly though: I don't see why simply submitting someone else's CSV file - shared publicly or privately - should lead to any kind of reward/recognition at all. It reminds me of this article by Charlie Brooker.)
\n\nThis would not address the people who blended or made tweaks to the CSV... but in the past nothing at all is done to fix last minute sharing. Merging implicit teams would be a small step up from that.
", "rawMarkdown": "> Please continue to post your thoughts and suggests in this thread. Your frank feedback is very much appreciated.\n\nOne small suggestion:\n\nWith data this size, tied scores are nearly certain to be identical CSVs.\n\nBreaking ties by time of submission becomes less meaningful/appropriate the more people submit a publicly shared CSV.\n\nWhy not group identical entries into one ***implicit team*** rank?\n\nIf you keep the original number of teams for medals calculations (as in two stage competitions) that would move some teams upwards into the medals zones. (Actually it may reward some admirable people - those who worked hard and gained an honest score just a hair below the shared solution but still did not use it.)\n\nPerhaps even dilute the points of the new merged teams in the same way actual teams are penalized by size. (More frankly though: I don't see why simply submitting someone else's CSV file - shared publicly or privately - should lead to any kind of reward/recognition at all. It reminds me of [this article by Charlie Brooker][1].)\n\nThis would not address the people who blended or made tweaks to the CSV... but in the past nothing at all is done to fix last minute sharing. *Merging implicit teams* would be a small step up from that.\n\n [1]: https://www.theguardian.com/commentisfree/2007/aug/06/comment.comment\n\n", "votes": 6 }, { "id": 325658, "postDate": "2018-05-08T17:08:58.043Z", "content": "That is a reasonable move, people who having exactly same submission count as one team and they would get few points. I think kaggle team needs to do something to those identical submissions, otherwise it could easily happen again when someone else wants to have some fun (like Dirk) and creates a new account , sharing high-rank solution at the final day..... This would be extremely discouraging for majority of competitors, if no penalty applied on it.
", "rawMarkdown": "That is a reasonable move, people who having exactly same submission count as one team and they would get few points. I think kaggle team needs to do something to those identical submissions, otherwise it could easily happen again when someone else wants to have some fun (like Dirk) and creates a new account , sharing high-rank solution at the final day..... This would be extremely discouraging for majority of competitors, if no penalty applied on it.", "votes": 3 }, { "id": 326480, "postDate": "2018-05-09T19:48:45.367Z", "content": "If kaggle stays like this, let the copies win, then most of the medals and ranks are valueless.\nThink about that some new platforms focused on truely fair competitions come out. Kaggle is ruining itself.
", "rawMarkdown": "If kaggle stays like this, let the copies win, then most of the medals and ranks are valueless.\nThink about that some new platforms focused on truely fair competitions come out. Kaggle is ruining itself.", "votes": 2 }, { "id": 326517, "postDate": "2018-05-09T21:20:44.323Z", "content": "Are there already some new platforms?
", "rawMarkdown": "Are there already some new platforms?" }, { "id": 326518, "postDate": "2018-05-09T21:21:50.680Z", "content": "Agreed. Kernels should become public only AFTER competition maybe.
", "rawMarkdown": "Agreed. Kernels should become public only AFTER competition maybe. ", "votes": -1 }, { "id": 326521, "postDate": "2018-05-09T21:28:52.513Z", "content": "Kernels are very helpful for beginners and sharing contributions. I don't agree with the idea that they should not be public, however they should be limited the final week of competition.
", "rawMarkdown": "Kernels are very helpful for beginners and sharing contributions. I don't agree with the idea that they should not be public, however they should be limited the final week of competition." }, { "id": 326543, "postDate": "2018-05-09T22:32:29.967Z", "content": "@Jihye topcoder has a data science channel.
", "rawMarkdown": "@Jihye topcoder has a data science channel." }, { "id": 326545, "postDate": "2018-05-09T22:38:02.520Z", "content": "Just block the result csv. I learn from kernels, not csv. What is the csv using for? To prove the kernel is true?
", "rawMarkdown": "Just block the result csv. I learn from kernels, not csv. What is the csv using for? To prove the kernel is true? " }, { "id": 326549, "postDate": "2018-05-09T22:47:31.197Z", "content": "@Jihye and NUMERAI
", "rawMarkdown": "@Jihye and NUMERAI" } ] }, { "id": 324772, "postDate": "2018-05-07T23:44:51.607Z", "content": "I worked hard for weeks to get score 0.9810, hoping to win my first silver medal, or at least a bronze medal
\n\nnow I just want to cry...
", "rawMarkdown": "I worked hard for weeks to get score 0.9810, hoping to win my first silver medal, or at least a bronze medal\n\nnow I just want to cry...", "votes": 29 }, { "id": 324392, "postDate": "2018-05-07T16:56:54.173Z", "content": "Spent many weeks and weekends literally and was in the top 5% until late in the evening. Now I see that I'm in the border of loosing my first medal that I was dreaming from the last 2 months.. \nNot sure what to say :( \nWish some one can understand the pain!!!
\n\nSamrat : i understand your position . i have felt the same way . let's hope for the best in PB.
", "rawMarkdown": "Samrat : i understand your position . i have felt the same way . let's hope for the best in PB.", "votes": 1 }, { "id": 324418, "postDate": "2018-05-07T17:22:29.510Z", "content": "@mayanksoni The huge difference will definitely push me down... In 1 hour I dropped almost 130 positions..
", "rawMarkdown": "@mayanksoni The huge difference will definitely push me down... In 1 hour I dropped almost 130 positions.. ", "votes": 1 }, { "id": 324424, "postDate": "2018-05-07T17:26:20.973Z", "content": "Whatever you choose now, you should be proud of what you accomplished with honest effort. In my opinion, what you learn from putting in that effort is much more valuable than your competition placement - both in terms of your broader ML/data science skillset and your ability to succeed in future kaggle competitions.
\n\nI hope that you won't feel discouraged from competing in future competitions - I'm sure that your results will only get better and better. In my first competition (Instacart), I crawled into the top 10% before plummeting when 50th place decided to open source their solution in the last week. But I learned a lot there and it helped me in more recent competitions. Kaggle success is a marathon, not a sprint.
", "rawMarkdown": "Whatever you choose now, you should be proud of what you accomplished with honest effort. In my opinion, what you learn from putting in that effort is much more valuable than your competition placement - both in terms of your broader ML/data science skillset and your ability to succeed in future kaggle competitions.\n\nI hope that you won't feel discouraged from competing in future competitions - I'm sure that your results will only get better and better. In my first competition (Instacart), I crawled into the top 10% before plummeting when 50th place decided to open source their solution in the last week. But I learned a lot there and it helped me in more recent competitions. Kaggle success is a marathon, not a sprint. ", "votes": 8 }, { "id": 324713, "postDate": "2018-05-07T22:45:48.277Z", "content": "Samrat, when I look at top kagglers, I understand they are not here for one competition but they have done many... and will continue to do many... At the end of the day, the score is less important than your real ability to solve thanks to machine learning a problem. This is what is going to be useful in your day to day job... Not the medals that are just good for flaunting one's ego
", "rawMarkdown": "Samrat, when I look at top kagglers, I understand they are not here for one competition but they have done many... and will continue to do many... At the end of the day, the score is less important than your real ability to solve thanks to machine learning a problem. This is what is going to be useful in your day to day job... Not the medals that are just good for flaunting one's ego", "votes": 3 }, { "id": 324762, "postDate": "2018-05-07T23:33:16.907Z", "rawMarkdown": "", "isDeleted": true }, { "id": 324854, "postDate": "2018-05-08T00:43:13.933Z", "content": "Public leaderboard justice for you. Congratulations.
", "rawMarkdown": "Public leaderboard justice for you. Congratulations.", "votes": 3 } ] }, { "id": 324658, "postDate": "2018-05-07T22:00:59.230Z", "content": "This is my first competition.
\n\nMy goal is to earn a bronze medal. The new kernel pushed my submission, and others, out of medal range.
\n\nThe best medal winning strategy, currently, is to copy and submit the highest scoring public kernel.
\n\nI've decided to submit my own solution, rather than the public model.
\n\nI hope that there is a policy change to address game breaking kernel submissions in the last days of the competition.
\n\nI quite like the Kaggle platform and community, but it's hard to justify spending weeks on a problem to compete with copy and paste.
", "rawMarkdown": "This is my first competition. \n\nMy goal is to earn a bronze medal. The new kernel pushed my submission, and others, out of medal range.\n\nThe best medal winning strategy, currently, is to copy and submit the highest scoring public kernel.\n\nI've decided to submit my own solution, rather than the public model. \n\nI hope that there is a policy change to address game breaking kernel submissions in the last days of the competition. \n\nI quite like the Kaggle platform and community, but it's hard to justify spending weeks on a problem to compete with copy and paste.", "votes": 22, "replies": [ { "id": 324680, "postDate": "2018-05-07T22:15:30.360Z", "content": "Hey, maybe you'll beat his model on the private LB. This doesn't happen for most competitions.
", "rawMarkdown": "Hey, maybe you'll beat his model on the private LB. This doesn't happen for most competitions.", "votes": 2 }, { "id": 324850, "postDate": "2018-05-08T00:41:58.313Z", "content": "I'm actually in the same position and feel the same way
", "rawMarkdown": "I'm actually in the same position and feel the same way", "votes": 3 } ] }, { "id": 324220, "postDate": "2018-05-07T12:26:48.227Z", "content": "I fully agree, and I wrote a similar post in Quora competition, see https://www.kaggle.com/c/quora-question-pairs/discussion/33801
\n\nI bet others have also written about the same issue in other competitions.
\n\nTo be clear, I am not against sharing high value kernels. I am against sharing high value kernels near the end of a competition. I personally would like that no new kernel sharing is possible in the last week.
", "rawMarkdown": "I fully agree, and I wrote a similar post in Quora competition, see https://www.kaggle.com/c/quora-question-pairs/discussion/33801\n\nI bet others have also written about the same issue in other competitions.\n\nTo be clear, I am not against sharing high value kernels. I am against sharing high value kernels near the end of a competition. I personally would like that no new kernel sharing is possible in the last week.", "votes": 22, "replies": [ { "id": 324227, "postDate": "2018-05-07T12:49:57.737Z", "content": "Yes, this topic comes up at nearly every competition deadline. My question is: what prevents you from blending with \"high value kernels\"? Just save a couple of submits before the final deadline and you are good.
", "rawMarkdown": "Yes, this topic comes up at nearly every competition deadline. My question is: what prevents you from blending with \"high value kernels\"? Just save a couple of submits before the final deadline and you are good. ", "votes": -5 }, { "id": 324234, "postDate": "2018-05-07T13:08:03.237Z", "content": "\n\n\nwhat prevents you from blending with \"high value kernels\"?
\n
My job ;)
\n\nOther reasons can be exams, travel, family events, whatever that can prevent you from working on a Kaggle competition for few days in a row.
\n\nWhatever the reason, it may happen that you cannot work on a competition the last day. If someone shares something valuable when one cannot react to it, then it is unfair IMHO.
\n\nFor me, the only way to counter it is to take a day off, like I'm doing today. But this is not always doable.
\n\nBut your question is valuable. That's why I am not against sharing in general, as one can always try to benefit from what has been shared. I guess this is your point, and I agree with it.
", "rawMarkdown": "> what prevents you from blending with \"high value kernels\"? \n\nMy job ;) \n\nOther reasons can be exams, travel, family events, whatever that can prevent you from working on a Kaggle competition for few days in a row. \n\nWhatever the reason, it may happen that you cannot work on a competition the last day. If someone shares something valuable when one cannot react to it, then it is unfair IMHO.\n\nFor me, the only way to counter it is to take a day off, like I'm doing today. But this is not always doable.\n\nBut your question is valuable. That's why I am not against sharing in general, as one can always try to benefit from what has been shared. I guess this is your point, and I agree with it.\n", "votes": 12 }, { "id": 324249, "postDate": "2018-05-07T13:29:25.733Z", "content": "People who do Kaggle full time (as myself) always have an advantage over those who also have a job. Some people, I assume, work in teams from one Kaggle profile which gives even more advantage. The problem of allocating your free time is related to opportunity costs. And it is totally irrelevant to the question of sharing the high-value kernels at the end of the competition.
", "rawMarkdown": "People who do Kaggle full time (as myself) always have an advantage over those who also have a job. Some people, I assume, work in teams from one Kaggle profile which gives even more advantage. The problem of allocating your free time is related to opportunity costs. And it is totally irrelevant to the question of sharing the high-value kernels at the end of the competition. ", "votes": 1 }, { "id": 324251, "postDate": "2018-05-07T13:31:04.847Z", "content": "\n\n\nPavel Pleskov wrote
\n \n\n\nYes, this topic comes up at nearly every competition deadline. My question is: what prevents you from blending with \"high value kernels\"? Just save a couple of submits before the final deadline and you are good.
\n
Computational resources ;)
", "rawMarkdown": "\n> **Pavel Pleskov wrote**\n> \n> > Yes, this topic comes up at nearly every competition deadline. My question is: what prevents you from blending with \"high value kernels\"? Just save a couple of submits before the final deadline and you are good. \n\nComputational resources ;)" }, { "id": 324254, "postDate": "2018-05-07T13:36:31.170Z", "content": "Blending someone else's work seems to be against the spirit of the competition and the ranking system. You can achieve a very high rank without knowing much about coding or machine learning. Doesn't seem right to me.
", "rawMarkdown": "Blending someone else's work seems to be against the spirit of the competition and the ranking system. You can achieve a very high rank without knowing much about coding or machine learning. Doesn't seem right to me.", "votes": 6 }, { "id": 324258, "postDate": "2018-05-07T13:46:55.193Z", "content": "@Pavel
\n\n\n\n\nSome people, I assume, work in teams from one Kaggle profile which gives even more advantage.
\n
I agree with you on that one. For the rest, let's agree that we disagree ;)
", "rawMarkdown": "@Pavel\n\n> Some people, I assume, work in teams from one Kaggle profile which gives even more advantage. \n\nI agree with you on that one. For the rest, let's agree that we disagree ;)", "votes": 3 }, { "id": 324265, "postDate": "2018-05-07T13:57:28.220Z", "content": "@Matthew I would love to see someone from top-100 (which is a very high rank for me) without knowing much about coding or machine learning :) IMHO getting the best possible score is indeed the spirit of the competition. What is wrong with finding creative ways to do it based on other people work? Isn't it how teamwork goes? By the way, have you ever tried to blend kernels and get the high score on a private leaderboard? Not that easy to do it consistently, trust me
", "rawMarkdown": "@Matthew I would love to see someone from top-100 (which is a very high rank for me) without knowing much about coding or machine learning :) IMHO getting the best possible score is indeed the spirit of the competition. What is wrong with finding creative ways to do it based on other people work? Isn't it how teamwork goes? By the way, have you ever tried to blend kernels and get the high score on a private leaderboard? Not that easy to do it consistently, trust me", "votes": 5 }, { "id": 324269, "postDate": "2018-05-07T14:03:32.927Z", "content": "Different TimeZone , Work that pays , Family ...
", "rawMarkdown": "Different TimeZone , Work that pays , Family ...", "votes": 2 }, { "id": 324271, "postDate": "2018-05-07T14:04:13.343Z", "content": "@Kevin there is always this complaint about lack of resources. Not enough time, experience, GPUs, cores, RAM... Yet some people still win without having all of that and not making excuses.
", "rawMarkdown": "@Kevin there is always this complaint about lack of resources. Not enough time, experience, GPUs, cores, RAM... Yet some people still win without having all of that and not making excuses." }, { "id": 324272, "postDate": "2018-05-07T14:05:32.807Z", "content": "right now you can run a public kernel and score in the top 150 without knowing anything about machine learning. @Pavel you should be upset about that because it makes your high ranking virtually meaningless
", "rawMarkdown": "right now you can run a public kernel and score in the top 150 without knowing anything about machine learning. @Pavel you should be upset about that because it makes your high ranking virtually meaningless", "votes": 1 }, { "id": 324281, "postDate": "2018-05-07T14:17:23.970Z", "content": "Right tail of the distribution never gets upset by the left tail :)\nPS: what if I tell you that there is also a private LB...
", "rawMarkdown": "Right tail of the distribution never gets upset by the left tail :)\nPS: what if I tell you that there is also a private LB..." }, { "id": 324283, "postDate": "2018-05-07T14:23:00.967Z", "content": "@Mathew, let's see the private LB before we can discuss ranks. Also, what is the public kernel that makes you in top 150 on the public LB? It needs to score at least 0.9811, and I don't remember seeing one.
", "rawMarkdown": "@Mathew, let's see the private LB before we can discuss ranks. Also, what is the public kernel that makes you in top 150 on the public LB? It needs to score at least 0.9811, and I don't remember seeing one." }, { "id": 324289, "postDate": "2018-05-07T14:29:13.047Z", "content": "So you're not upset if a left-tailer can masquerade as a right-tailer by using quality public kernels posted on the last day? If you score in the top 150 in every competition by simply running a good kernel on the last day, wouldn't you be ranked among the top 100 overall? I think the ranking system is flawed but I am not sure how to fix it.
", "rawMarkdown": "So you're not upset if a left-tailer can masquerade as a right-tailer by using quality public kernels posted on the last day? If you score in the top 150 in every competition by simply running a good kernel on the last day, wouldn't you be ranked among the top 100 overall? I think the ranking system is flawed but I am not sure how to fix it.", "votes": 1 }, { "id": 324290, "postDate": "2018-05-07T14:31:47.187Z", "content": "@CPMP He is referring to this one.
", "rawMarkdown": "@CPMP He is referring to [this one][1].\n\n\n [1]: https://www.kaggle.com/midi303/entire-dataset-lb-0-9811-lightgbm", "votes": 2 }, { "id": 324291, "postDate": "2018-05-07T14:32:11.217Z", "content": "Yes there is one that scores 0.9811. People have been complaining about it this morning.
", "rawMarkdown": "Yes there is one that scores 0.9811. People have been complaining about it this morning. ", "votes": 2 }, { "id": 324302, "postDate": "2018-05-07T14:48:57.157Z", "content": "My bad, I didn't see it. Will look at it right now!
\n\nThis sharing so late is silly.
\n\nEdited: no low hanging fruit that I can reuse in it, will stick to my plan for the next few hours. That's why sharing so late is silly, people don't have time to react to it. I feel for all those who put hard work and will be displaced by people who happen to have machines large enough to run this before competition ends. The only counterpart is that this kernel may overfit to the public LB (like my code now that I think of it ... )
", "rawMarkdown": "My bad, I didn't see it. Will look at it right now! \n\nThis sharing so late is silly.\n\nEdited: no low hanging fruit that I can reuse in it, will stick to my plan for the next few hours. That's why sharing so late is silly, people don't have time to react to it. I feel for all those who put hard work and will be displaced by people who happen to have machines large enough to run this before competition ends. The only counterpart is that this kernel may overfit to the public LB (like my code now that I think of it ... )", "votes": 1 }, { "id": 324306, "postDate": "2018-05-07T14:52:51.717Z", "content": "\n\n\nCPMP wrote
\n \nI agree with you on that one. For the rest, let's agree that we disagree ;)
\n
'Men in black' is quite enjoying movie
", "rawMarkdown": "\n> **CPMP wrote**\n\n> I agree with you on that one. For the rest, let's agree that we disagree ;)\n\n'Men in black' is quite enjoying movie ", "votes": 1 }, { "id": 324307, "postDate": "2018-05-07T14:53:42.430Z", "content": "And you even don't have to run it ...
", "rawMarkdown": "And you even don't have to run it ...", "votes": 3 }, { "id": 324361, "postDate": "2018-05-07T16:18:50.977Z", "content": "@Matthew according to my ranking score tool https://www.kaggle.com/ppleskov/ranking-comparison-tool 150th place gives you around 1500 points, so to reach the top-100 score of 38000 you will need to perform this trick in 25 competitions. It is highly unlikely to do so in a reasonable time since not in every competition somebody shares high results on the last day and there are around 20 competitions per year. I'm not even talking about reaching the very top of 180000 points. So yes, it does not bother me at all.
", "rawMarkdown": "@Matthew according to my ranking score tool https://www.kaggle.com/ppleskov/ranking-comparison-tool 150th place gives you around 1500 points, so to reach the top-100 score of 38000 you will need to perform this trick in 25 competitions. It is highly unlikely to do so in a reasonable time since not in every competition somebody shares high results on the last day and there are around 20 competitions per year. I'm not even talking about reaching the very top of 180000 points. So yes, it does not bother me at all. ", "votes": 4 }, { "id": 324375, "postDate": "2018-05-07T16:28:53.260Z", "content": "Wow. Someone with no knowledge of machine learning could theoretically be ranked in the top 100 on Kaggle within just a few years. That bothers me and I haven't invested time in acquiring a high rank. I think it's just unfair to those that have invested a lot of time to signal their skills with a high rank.
", "rawMarkdown": "Wow. Someone with no knowledge of machine learning could theoretically be ranked in the top 100 on Kaggle within just a few years. That bothers me and I haven't invested time in acquiring a high rank. I think it's just unfair to those that have invested a lot of time to signal their skills with a high rank. ", "votes": 2 }, { "id": 324383, "postDate": "2018-05-07T16:42:58.563Z", "content": "@Mathew, The issue does not impact top performers, see Pavel's reaction above. I agree with him that there is no way someone can make it in top 150 in the global ranking system just by using public kernel outputs. No way.
\n\nThat's not why this sharing is an issue. It is an issue because of all whose ranks will be negatively impacted in this competition may not want to invest time in the next competition.
", "rawMarkdown": "@Mathew, The issue does not impact top performers, see Pavel's reaction above. I agree with him that there is no way someone can make it in top 150 in the global ranking system just by using public kernel outputs. No way.\n\nThat's not why this sharing is an issue. It is an issue because of all whose ranks will be negatively impacted in this competition may not want to invest time in the next competition.", "votes": 5 }, { "id": 324398, "postDate": "2018-05-07T17:03:57.547Z", "content": "@CPMP I am new so don't know how often this happens. So this late posting doesn't happen often? Others seem to think it happens all the time. In any event, it would be easy to stop if Kaggle just blocked new kernel posts a few days before the end of a competition.
", "rawMarkdown": "@CPMP I am new so don't know how often this happens. So this late posting doesn't happen often? Others seem to think it happens all the time. In any event, it would be easy to stop if Kaggle just blocked new kernel posts a few days before the end of a competition." }, { "id": 324444, "postDate": "2018-05-07T17:52:30.817Z", "content": "Check out the excellent Scripty McScriptface the Lazy Kaggler analysis by dune_dweller.
\n\nFor the first 15 months or so of Scripts/Kernels, submitting the best public script could get you to\n13881 points, at that time good for 449th globally...
\n\nIt would be fantastic if Kaggle updated the Meta Kaggle data set & that analysis kernel could be re-run on all the competitions to date...
", "rawMarkdown": "Check out the excellent [Scripty McScriptface the Lazy Kaggler][1] analysis by [dune_dweller][2].\n\nFor the first 15 months or so of Scripts/Kernels, submitting the best public script could get you to\n13881 points, at that time good for 449th globally...\n\nIt would be fantastic if Kaggle updated the [Meta Kaggle][3] data set & that analysis kernel could be re-run on all the competitions to date...\n\n\n [1]: https://www.kaggle.com/dvasyukova/scripty-mcscriptface-the-lazy-kaggler\n [2]: https://www.kaggle.com/dvasyukova\n [3]: https://www.kaggle.com/kaggle/meta-kaggle\n\n", "votes": 3 }, { "id": 324454, "postDate": "2018-05-07T17:56:32.990Z", "content": "Yeah DuneWeller is awesome - the good news from this was that at least they removed the automated submission button - zoiks! If that was still there my dog would come in the top 500
", "rawMarkdown": "Yeah DuneWeller is awesome - the good news from this was that at least they removed the automated submission button - zoiks! If that was still there my dog would come in the top 500", "votes": 2 }, { "id": 324677, "postDate": "2018-05-07T22:13:06.147Z", "content": "Agree that this kind of late kernel does not harm top performers. This is bad that it is shared at the last minute but it is usefull for people like me who are at the learning stage to see imagination of other people.
", "rawMarkdown": "Agree that this kind of late kernel does not harm top performers. This is bad that it is shared at the last minute but it is usefull for people like me who are at the learning stage to see imagination of other people.", "votes": 2 }, { "id": 325138, "postDate": "2018-05-08T06:54:05.007Z", "content": "Useful to preserve the plaintext names of both the submitter and kernel, since the kernel was removed and the submitter seems to have been disqualified:
\n\nKernel: www.kaggle.com/midi303/entire-dataset-lb-0-9811-lightgbm
\n\nSubmitter: Dirk www.kaggle.com/midi303
\n\n(I'm pretty sure this is still public knowledge, since it's in each of our inboxes if you subscribed to email delivery)
", "rawMarkdown": "Useful to preserve the plaintext names of both the submitter and kernel, since the kernel was removed and the submitter seems to have been disqualified:\n\n**Kernel:** www.kaggle.com/midi303/entire-dataset-lb-0-9811-lightgbm\n\n**Submitter:** Dirk " } ] }, { "id": 324831, "postDate": "2018-05-08T00:31:09.353Z", "content": "It is so bad to see silver region from 153-171 all share one kernel \"Merge and avg.\" and gain silver medal with less than 10 submissions. I suggest committee to cancel grade for all participants with exactly same submission. This file is super large and exact same submission means plagirism
", "rawMarkdown": "It is so bad to see silver region from 153-171 all share one kernel \"Merge and avg.\" and gain silver medal with less than 10 submissions. I suggest committee to cancel grade for all participants with exactly same submission. This file is super large and exact same submission means plagirism", "votes": 19, "replies": [ { "id": 324945, "postDate": "2018-05-08T02:03:18.237Z", "content": "This is the actual definition of harmful plagiarism, which shouldnāt be allowed on Kaggle as it isnāt allowed in any other academic setting.
", "rawMarkdown": "This is the actual definition of harmful plagiarism, which shouldnāt be allowed on Kaggle as it isnāt allowed in any other academic setting.", "votes": 4 }, { "id": 325003, "postDate": "2018-05-08T03:08:11.993Z", "content": "Well said Sun Zhiaho
", "rawMarkdown": "Well said Sun Zhiaho" }, { "id": 325268, "postDate": "2018-05-08T09:12:43.170Z", "content": "Plagiarism is a heavy term and I don't like seeing it thrown around lightly. In Kaggle it is within the rules to use public kernel submissions, so this is not plagiarism since plagiarism revolves around \"stealing\" someone else's work and claiming it as your own. Under Kaggle rules, submitting a public kernel submission is within the rules so this is by definition not plagiarism.
\n\nI would like to see a solution that properly resolves this as much as anyone else on here, but throwing terms and accusations around is not conductive to the discussion.
", "rawMarkdown": "Plagiarism is a heavy term and I don't like seeing it thrown around lightly. In Kaggle it is within the rules to use public kernel submissions, so this is not plagiarism since plagiarism revolves around \"stealing\" someone else's work and claiming it as your own. Under Kaggle rules, submitting a public kernel submission is within the rules so this is by definition not plagiarism.\n\nI would like to see a solution that properly resolves this as much as anyone else on here, but throwing terms and accusations around is not conductive to the discussion.", "votes": 1 }, { "id": 325606, "postDate": "2018-05-08T16:12:43.070Z", "content": "Anthony, I agree that blind accusations are wrong, but exactly resubmitting the output of other peoples' work without even running the kernel itself does seem wrong and appears to be very similar to plagiarism. Plagiarism, as in the practice of taking someone else's work or ideas (kernels) and passing them off as one's own (on the LB).
", "rawMarkdown": "Anthony, I agree that blind accusations are wrong, but exactly resubmitting the output of other peoples' work without even running the kernel itself does seem wrong and appears to be very similar to plagiarism. Plagiarism, as in the practice of taking someone else's work or ideas (kernels) and passing them off as one's own (on the LB)." } ] }, { "id": 325065, "postDate": "2018-05-08T04:43:33.110Z", "content": "It seems as results stand currently, it's at least 30% of medals are for exact copied work without any additions at all. \nPerhaps even more people leveraged late solutions in their blends. This system promotes people to think less, not more. I wonder how many of these folks even understand what they copy - we don't have a way to tell.
\n\nI wonder if people who submit these solutions really want to showcase such a result on resume... This is such a shame show. To discourage it, I could suggest Kaggle to
\n\nI would fully agree for restricted sharing in the end of the competition as well. Typically people focus on working on their solutions at this time, not everyone has a chance to react to public kernel in a short period of time even if they wanted to.
\n\nI think also it makes sense for Kaggle to retro-actively adjust results of this competition itself to help recognize the modeling work that people put in.
\n\nHope this helps figure out a good approach that promotes excellence and collaboration, while giving credit where it's due.
", "rawMarkdown": "It seems as results stand currently, it's at least 30% of medals are for exact copied work without any additions at all. \nPerhaps even more people leveraged late solutions in their blends. This system promotes people to think less, not more. I wonder how many of these folks even understand what they copy - we don't have a way to tell.\n\nI wonder if people who submit these solutions really want to showcase such a result on resume... This is such a shame show. To discourage it, I could suggest Kaggle to \n\n* only give credit to creator of the first published solution among identical submissions (even if it is not selected as final, might want to think a bit more of what we call identical); we should not allow identical solution to be selected for private leaderboard submission for people who copy it.\n* publish a few more decimals of the private score and highlight identical submission on leaderboard as duplicate, to make it even more obvious that it's exact same solutions being submitted\n\nI would fully agree for restricted sharing in the end of the competition as well. Typically people focus on working on their solutions at this time, not everyone has a chance to react to public kernel in a short period of time even if they wanted to. \n\nI think also it makes sense for Kaggle to retro-actively adjust results of this competition itself to help recognize the modeling work that people put in. \n\nHope this helps figure out a good approach that promotes excellence and collaboration, while giving credit where it's due.", "votes": 16, "replies": [ { "id": 325164, "postDate": "2018-05-08T07:23:23.557Z", "content": "Great idea! But people can blend anyway, there will be kernels of how to blend with different random seeds.
", "rawMarkdown": "Great idea! But people can blend anyway, there will be kernels of how to blend with different random seeds.", "votes": 2 } ] }, { "id": 324416, "postDate": "2018-05-07T17:21:25.490Z", "content": "Why would recruiters use kaggle now that these blinking kernels exist. Kaggle sort it out - you will lose hard working competitors too! When I wrote the kernels schmernels post I was inaccurate when I said start two weeks from the end - apologies ;(
", "rawMarkdown": "Why would recruiters use kaggle now that these blinking kernels exist. Kaggle sort it out - you will lose hard working competitors too! When I wrote the kernels schmernels post I was inaccurate when I said start two weeks from the end - apologies ;(", "votes": 15 }, { "id": 325412, "postDate": "2018-05-08T11:48:42.450Z", "content": "One thing to not forget - if that kernel hadn't been released - this was a very very successful competition overall - lots of scope for feature engineering that didn't have leakage!
", "rawMarkdown": "One thing to not forget - if ***that kernel*** hadn't been released - this was a very very successful competition overall - lots of scope for feature engineering that didn't have leakage!", "votes": 11 }, { "id": 324393, "postDate": "2018-05-07T16:58:12.113Z", "content": "It's very disappointing to see LB being destroyed in last hours. I am sure that Kaggle can't do anything about it because they will never know who will be posting what and when. But at least they can have some mechanism to allow community to punish such D***s flying around in every competition in last hours.
This is regular problem and needs a solution. I and many others strongly recommend having downvote option for kernels .
\n\nMany kagglers have already educated that novice kernel author but it seems that he is purely in the mood of destroying leaderboard and not taking down that kernel. So may be there is a need to have a Hall of Shame post or a kernel like this
", "rawMarkdown": "It's very disappointing to see LB being destroyed in last hours. I am sure that Kaggle can't do anything about it because they will never know who will be posting what and when. But at least they can have some mechanism to allow community to punish such `D***s` flying around in every competition in last hours. \n\nThis is regular problem and needs a solution. I and many others strongly recommend having **downvote option for kernels** . \n\nMany kagglers have already educated that novice kernel author but it seems that he is purely in the mood of destroying leaderboard and not taking down that kernel. So may be there is a need to have a Hall of Shame post or a kernel like [this][1] \n\n\n [1]: https://www.kaggle.com/carloshuertas/kaggle-users-with-most-awful-forum-karma", "votes": 12, "replies": [ { "id": 324681, "postDate": "2018-05-07T22:15:48.307Z", "content": "All, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel. I understand this can be seen like cheating for people who are using this kernel to increase their score but for top performers, they are way above this 0.9811 so this should not be such an issue on the kaggle forum...
", "rawMarkdown": "All, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel. I understand this can be seen like cheating for people who are using this kernel to increase their score but for top performers, they are way above this 0.9811 so this should not be such an issue on the kaggle forum... ", "votes": -4 }, { "id": 324686, "postDate": "2018-05-07T22:18:42.273Z", "content": "\n\n\neric wrote
\n \n\n\nAll, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel. I understand this can be seen like cheating for people who are using this kernel to increase their score but for top performers, they are way above this 0.9811 so this should not be such an issue on the kaggle forum...
\n
It's not ironic at all. Many of us care about the broader community beyond ourselves and want to see diligent, honest work rewarded over last minute button pressing. Also, many of us have been in this exact position in the past and understand how frustrating it is, so can empathize.
", "rawMarkdown": "\n> **eric wrote**\n> \n> > All, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel. I understand this can be seen like cheating for people who are using this kernel to increase their score but for top performers, they are way above this 0.9811 so this should not be such an issue on the kaggle forum... \n\nIt's not ironic at all. Many of us care about the broader community beyond ourselves and want to see diligent, honest work rewarded over last minute button pressing. Also, many of us have been in this exact position in the past and understand how frustrating it is, so can empathize. ", "votes": 12 }, { "id": 324701, "postDate": "2018-05-07T22:34:49.127Z", "content": "Joe Eddy, thanks for being so compassionate about poor kagglers like me. But I still believe the real issue is to remember that Kaggle has been created to make learning data science and machine learning a game and that in any game the most important stuff at least for me is to participate and have some fun. Then, if you win, this is even better but winning should not be seen as the ultimate goal
\n\nAnd I cannot resist to quote a famous French countryman\n\"The most important thing in the Olympic Games is not winning but taking part; the essential thing in life is not conquering but fighting well.\" - Pierre de Coubertin
", "rawMarkdown": "Joe Eddy, thanks for being so compassionate about poor kagglers like me. But I still believe the real issue is to remember that Kaggle has been created to make learning data science and machine learning a game and that in any game the most important stuff at least for me is to participate and have some fun. Then, if you win, this is even better but winning should not be seen as the ultimate goal\n\nAnd I cannot resist to quote a famous French countryman\n\"The most important thing in the Olympic Games is not winning but taking part; the essential thing in life is not conquering but fighting well.\" - Pierre de Coubertin", "votes": -7 }, { "id": 324728, "postDate": "2018-05-07T23:01:14.760Z", "content": "\n\n\neric wrote
\n \n\n\nAll, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel.
\n
When you're part of the community, you can feel the pain and would stand up naturally.
\n\nI have faced such last day disasters in some of my previous competitions and can totally relate how it feels when few people (ya, just few out of thousand participants) destroys the leaderboard.
", "rawMarkdown": "\n> **eric wrote**\n> \n> > All, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel. \n\nWhen you're part of the community, you can feel the pain and would stand up naturally. \n\nI have faced such last day disasters in some of my previous competitions and can totally relate how it feels when few people (ya, just few out of thousand participants) destroys the leaderboard. ", "votes": 3 }, { "id": 324732, "postDate": "2018-05-07T23:04:22.313Z", "content": "Thanks Pranav for this nice feedback! Understood then
", "rawMarkdown": "Thanks Pranav for this nice feedback! Understood then", "votes": 1 } ] }, { "id": 324358, "postDate": "2018-05-07T16:13:09.130Z", "content": "within last hour, many with <5 subs jumps to this score. Now imagine you were someone having 9810 last night and thought you could get your first silver medal.
", "rawMarkdown": "within last hour, many with <5 subs jumps to this score. Now imagine you were someone having 9810 last night and thought you could get your first silver medal.", "votes": 9 }, { "id": 325047, "postDate": "2018-05-08T03:59:36.907Z", "content": "as a new player of kaggle, I just want to cry. Very disappointed.
", "rawMarkdown": "as a new player of kaggle, I just want to cry. Very disappointed.", "votes": 8, "replies": [ { "id": 326548, "postDate": "2018-05-09T22:44:37.227Z", "content": "I feel the same. I feel maybe I came to a wrong place, or I should learn something shame rather than real knowledge if I want a medal.
", "rawMarkdown": "I feel the same. I feel maybe I came to a wrong place, or I should learn something shame rather than real knowledge if I want a medal." } ] }, { "id": 325027, "postDate": "2018-05-08T03:32:22.803Z", "content": "Very bad Kaggle experience this morning.
", "rawMarkdown": "Very bad Kaggle experience this morning.", "votes": 7 }, { "id": 324411, "postDate": "2018-05-07T17:17:50.487Z", "content": "It is interesting to see how many people are willing to use the infamous late kernel to obtain a now meaningless high rank in this competition. Maybe that should be the next competition: predict which Kagglers will make the late submission of someone else's work.
", "rawMarkdown": "It is interesting to see how many people are willing to use the infamous late kernel to obtain a now meaningless high rank in this competition. Maybe that should be the next competition: predict which Kagglers will make the late submission of someone else's work.", "votes": 8, "replies": [ { "id": 324539, "postDate": "2018-05-07T19:24:29.323Z", "content": "I like the idea. We will have training set as well :D
", "rawMarkdown": "I like the idea. We will have training set as well :D", "votes": 3 }, { "id": 324994, "postDate": "2018-05-08T02:52:09.337Z", "content": "I am looking forward to join you in that competition.
", "rawMarkdown": "I am looking forward to join you in that competition." } ] }, { "id": 324384, "postDate": "2018-05-07T16:44:40.627Z", "content": "It is worse now there are datasets - I have been warning about this for months - you just get a \"we can't do anything about it\" which is something I would fire someone over if they worked for me (I am a bad boss I suppose)
", "rawMarkdown": "It is worse now there are datasets - I have been warning about this for months - you just get a \"we can't do anything about it\" which is something I would fire someone over if they worked for me (I am a bad boss I suppose)", "votes": 8 }, { "id": 324425, "postDate": "2018-05-07T17:26:29.517Z", "content": "I think, that this topic has made 0.9811 kernel much more popular.
", "rawMarkdown": "I think, that this topic has made 0.9811 kernel much more popular.", "votes": 5 }, { "id": 324405, "postDate": "2018-05-07T17:10:04.963Z", "content": "I went to sleep in silver medal position, then I woke up and found that infamous kernel. I have made a significant effort and I was enjoying the competition, now this situation is very frustrating. Now ,as many competitors, I am at work and don't have the time and resources to train more models. So, kaggle user who want to use this kernel consider the fairness of the competition and don't use it please.
", "rawMarkdown": "I went to sleep in silver medal position, then I woke up and found that infamous kernel. I have made a significant effort and I was enjoying the competition, now this situation is very frustrating. Now ,as many competitors, I am at work and don't have the time and resources to train more models. So, kaggle user who want to use this kernel consider the fairness of the competition and don't use it please. ", "votes": 6 }, { "id": 324345, "postDate": "2018-05-07T15:54:26.423Z", "content": "Agree with you, investing big time and the last day you find high score Kernal where no time to learn from it or maybe you don't have submissions left to use it ! then hours of your afford vanish.
\n\nMy suggestion after the rules acceptance deadline to stop public Kernals because no much time to learn from it. OR at least stop upload input to public Kernal at that date
\n\nAlso regarding Kaggle removing scores of competitors because of cheating usually happen for unaware of rules to publish results primitively or share it with friend, I suggest Kaggle make a run to remove them at acceptance deadline, As people decide to invest more and more time on competition after it or could decide to rent a server and pay money then find score vanished by the end. Also more fair to have time to send appeal in case required.
", "rawMarkdown": "Agree with you, investing big time and the last day you find high score Kernal where no time to learn from it or maybe you don't have submissions left to use it ! then hours of your afford vanish.\n\n**My suggestion after the rules acceptance deadline to stop public Kernals because no much time to learn from it. OR at least stop upload input to public Kernal at that date**\n\nAlso regarding Kaggle removing scores of competitors because of cheating usually happen for unaware of rules to publish results primitively or share it with friend, I suggest Kaggle make a run to remove them at acceptance deadline, As people decide to invest more and more time on competition after it or could decide to rent a server and pay money then find score vanished by the end. Also more fair to have time to send appeal in case required.", "votes": 4 }, { "id": 324767, "postDate": "2018-05-07T23:38:30.560Z", "content": "I want to resign from this competition.
\n\nYes, the ranks matter (it is a competition!) and yes I m pissed that my participation is going to benefit last minute public blending competitors!
", "rawMarkdown": "I want to resign from this competition.\n\nYes, the ranks matter (it is a competition!) and yes I m pissed that my participation is going to benefit last minute public blending competitors!", "votes": 5 }, { "id": 324991, "postDate": "2018-05-08T02:47:41.170Z", "content": "Is it cheating or unethical? I am completely against the term cheating here. There's no rule in the competition that describes public sharing at the last moment is not allowed and will be termed as \"Cheater\"
\n\nCouple of cases -
\n\n\n\n\nSomebody uses the csv to upload. -- Is it against rule?\n Somebody blended with their model-- is it cheating?\n Somebody uses some of their features and used it ( like me) in their model. -- what would you call it?\n Somebody who did their blending at last moment with/without using this kernel's CSV. -- can it be tracked?
\n
Somehow I would term the entire episode as Unethical rather than cheating. I am not sure what Kaggle is going to do today but its unfair to remove all those who directly/indirectly uses the kernel. Rather as already mentioned , Kaggle should do something in future competition to stop the practice.
", "rawMarkdown": "Is it cheating or unethical? I am completely against the term **cheating** here. There's no rule in the competition that describes public sharing at the last moment is not allowed and will be termed as \"Cheater\"\n\nCouple of cases -\n> Somebody uses the csv to upload. -- Is it against rule?\n> Somebody blended with their model-- is it cheating?\n> Somebody uses some of their features and used it ( like me) in their model. -- what would you call it?\n> Somebody who did their blending at last moment with/without using this kernel's CSV. -- can it be tracked?\n\nSomehow I would term the entire episode as Unethical rather than cheating. I am not sure what Kaggle is going to do today but its unfair to remove all those who directly/indirectly uses the kernel. Rather as already mentioned , Kaggle should do something in future competition to stop the practice.", "votes": 3, "replies": [ { "id": 325000, "postDate": "2018-05-08T03:05:50.283Z", "content": "@ Subiksha Pal
\n\nYou used someone else's output. If you did that in a University you would get expelled/disbarred. \nIf you did that in a company you would be hit with a class action lawsuit for intellectual property theft. Perhaps you should bear that in mind before defending your position. What is really disturbing is how people are actually defending flagrant PLIAGARISM
\n\nI have no problems with blending or taking ideas from someone else or even last minute kernel releases but even that needs to be fair. You can't Float the ANSWERS around for everybody to use.
", "rawMarkdown": "@ Subiksha Pal \n\nYou used someone else's output. If you did that in a University you would get expelled/disbarred. \nIf you did that in a company you would be hit with a class action lawsuit for intellectual property theft. Perhaps you should bear that in mind before defending your position. What is really disturbing is how people are actually defending flagrant PLIAGARISM\n\nI have no problems with blending or taking ideas from someone else or even last minute kernel releases but even that needs to be fair. You can't Float the ANSWERS around for everybody to use. " }, { "id": 325013, "postDate": "2018-05-08T03:20:24.823Z", "content": "I am not defending here . We all are here talking about this fairness. But how to define that? My perception was to have some rule to confine the publication of the kernel at last minutes.These all are unethical practice that I am trying to mention and should be stopped.
", "rawMarkdown": "I am not defending here . We all are here talking about this fairness. But how to define that? My perception was to have some rule to confine the publication of the kernel at last minutes.These all are unethical practice that I am trying to mention and should be stopped." }, { "id": 325262, "postDate": "2018-05-08T09:06:17.700Z", "content": "@Shanth: I am as upset by this behavior as anyone, but this is in no way cheating since it broke no rules. It is not IP theft, since the guy shared the results publicly (and I'm sure kernels are published under some license, but I can't be bothered to check). In Universities you are explicitly not allowed to share exams/assignments, but in Kaggle this is encouraged by public kernels, so this is not a correct analogy as well.
\n\nIt is an ethical issue where users released competition-breaking content in the last second. Everything was within the rules.
", "rawMarkdown": "@Shanth: I am as upset by this behavior as anyone, but this is in no way cheating since it broke no rules. It is not IP theft, since the guy shared the results publicly (and I'm sure kernels are published under some license, but I can't be bothered to check). In Universities you are explicitly not allowed to share exams/assignments, but in Kaggle this is encouraged by public kernels, so this is not a correct analogy as well.\n\nIt is an ethical issue where users released competition-breaking content in the last second. Everything was within the rules.", "votes": 1 }, { "id": 325320, "postDate": "2018-05-08T09:58:54.017Z", "content": "@ Anthony Marakis -
\n\nLet's put this into context here a little bit. The problem we were working on needed extra computing power and memory optimizations to run the training models across the entire data set. Not everyone could do this for lack of resources. So, someone runs the model on all the data and understandably gets better results ( The 0.9811 kernel had features that were no different from the public kernels and got better results simply because it was using more data ) and decides to share the output file.
\n\nOthers, who have themselves not had access to such resources decide to USE THE OUTPUT FILE without even running a kernel and thats' OKAY? Hundreds of people UPLOAD THE SAME OUTPUT file and that's okay and that's NOT PLAGIARISM?
\n\nI don't think plagiarism is a heavy term. Plagiarism is when one copies someone else's work and passes it off as their own. Wouldn't this qualify ?\n :)
", "rawMarkdown": "@ Anthony Marakis - \n\nLet's put this into context here a little bit. The problem we were working on needed extra computing power and memory optimizations to run the training models across the entire data set. Not everyone could do this for lack of resources. So, someone runs the model on all the data and understandably gets better results ( The 0.9811 kernel had features that were no different from the public kernels and got better results simply because it was using more data ) and decides to share the output file. \n\nOthers, who have themselves not had access to such resources decide to USE THE OUTPUT FILE without even running a kernel and thats' OKAY? Hundreds of people UPLOAD THE SAME OUTPUT file and that's okay and that's NOT PLAGIARISM?\n\nI don't think plagiarism is a heavy term. Plagiarism is when one copies someone else's work and passes it off as their own. Wouldn't this qualify ?\n :)", "votes": -1 }, { "id": 325332, "postDate": "2018-05-08T10:11:20.990Z", "content": "I wholeheartedly agree that there is an issue (that's why I wrote this post :)). Actually, this is a huge issue. But plagiarism did not happen here. Plagiarism does not involve simple copying, but stealing. It is a type theft, which did not happen here.
\n\nThe only reason I don't want plagiarism to be thrown so easily around is because I know how terrible it can be for people involved. I was a part of a large community of writers, and there were cases where people made money plagiarizing (stealing licensed work and passing it off as their own) the work of others until they were caught and sued to oblivion. Even then, the damages to the writers were probably in the thousands of dollars, since they couldn't sell their own work, because someone else was claiming it was theirs (and takes a ton of time, money and effort to prove something is yours). It's a very nasty business, probably the nastiest in the creative community, and I don't want the term to start losing its meaning.
\n\nI prefer to see our efforts into fixing the actual issue, since this cannot continue. I am seriously considering not participating until this is fixed, even though I was one of the lucky ones who ended up right where they were before the shenanigans started.
", "rawMarkdown": "I wholeheartedly agree that there is an issue (that's why I wrote this post :)). Actually, this is a huge issue. But plagiarism did not happen here. Plagiarism does not involve simple copying, but stealing. It is a type theft, which did not happen here.\n\nThe only reason I don't want plagiarism to be thrown so easily around is because I know how terrible it can be for people involved. I was a part of a large community of writers, and there were cases where people made money plagiarizing (stealing licensed work and passing it off as their own) the work of others until they were caught and sued to oblivion. Even then, the damages to the writers were probably in the thousands of dollars, since they couldn't sell their own work, because someone else was claiming it was theirs (and takes a ton of time, money and effort to prove something is yours). It's a very nasty business, probably the nastiest in the creative community, and I don't want the term to start losing its meaning.\n\nI prefer to see our efforts into fixing the actual issue, since this cannot continue. I am seriously considering not participating until this is fixed, even though I was one of the lucky ones who ended up right where they were before the shenanigans started.", "votes": 2 } ] }, { "id": 325334, "postDate": "2018-05-08T10:12:31.390Z", "content": "I would say no sharing of public kernels that can achieve top 20% score. It was disappointing I was in the top 6% last night. Got up to find 300 new people on top of me with the same score!! ARGH!!
", "rawMarkdown": "I would say no sharing of public kernels that can achieve top 20% score. It was disappointing I was in the top 6% last night. Got up to find 300 new people on top of me with the same score!! ARGH!!", "votes": 3 }, { "id": 324655, "postDate": "2018-05-07T21:59:56.930Z", "content": "How about we post a 0.9832 solution? (but with random noise in the private hours)
\nJust kidding...
There were talks about blends spoiling the community, blending attempts at last minute of this competition has made this even worse.
", "rawMarkdown": "There were talks about blends spoiling the community, blending attempts at last minute of this competition has made this even worse.", "votes": 4 }, { "id": 324559, "postDate": "2018-05-07T19:40:31.953Z", "content": "For all those punters wishing to sleep tonight click on the option to unsubscribe ;)
", "rawMarkdown": "For all those punters wishing to sleep tonight click on the option to unsubscribe ;)", "votes": 4 }, { "id": 324522, "postDate": "2018-05-07T19:11:26.350Z", "content": "Can I withdraw from the competition?
\n\nI already have two dark spots on my portfolio:
\n\nSanta Gift Matching Challenge - Infinite Probabilistic Improver was published when I was on holiday
\n\nMercedes-Benz Greener Manufacturing - for most of us it seems the Manatees decided the place on LB
\n\nNow this.
", "rawMarkdown": "Can I withdraw from the competition?\n\nI already have two dark spots on my portfolio:\n\nSanta Gift Matching Challenge - Infinite Probabilistic Improver was published when I was on holiday\n\nMercedes-Benz Greener Manufacturing - for most of us it seems the [Manatees][1] decided the place on LB\n\nNow this.\n\n\n[1]: http://southpark.wikia.com/wiki/Manatees", "votes": 4 }, { "id": 324456, "postDate": "2018-05-07T18:00:27.043Z", "content": "It seems there are already a lot of good suggestions here. My question is how to get Kaggle officials' attention and act on it. Personally I think it is bad for Kaggle as a company as well for various reasons:
\n\n1) The most attractive part of Kaggle is the sheer amount of excellent data scientists that are actively learning, sharing and developing methods. Last-minute high-score kernel/results sharing will definitely disappoint and drive away many good and hard working contributors, which I don't think Kaggle wanna see.
\n\n2) My guesses are part of Kaggle's business model is for companies to recruit from Kaggle platform. If things like this happen, where the ranking of a competition become rather meaningless, I don't think recruiter will see any value here. As a matter of fact, I found more and more recruiters and people in data science community think Kaggle is just for competition but not real data science experience/projects. I think that is bad for the company as well.
\n\nIt takes a long time and effort to build a platform this great and have so many wonderful data scientists (I am relatively new, but already learnt a lot from many discussions posts and kernels). I hope there is a way to keep improving it, keep all the great people around and keep the competition rankings more meaningful.
", "rawMarkdown": "It seems there are already a lot of good suggestions here. My question is how to get Kaggle officials' attention and act on it. Personally I think it is bad for Kaggle as a company as well for various reasons:\n\n1) The most attractive part of Kaggle is the sheer amount of excellent data scientists that are actively learning, sharing and developing methods. Last-minute high-score kernel/results sharing will definitely disappoint and drive away many good and hard working contributors, which I don't think Kaggle wanna see. \n\n2) My guesses are part of Kaggle's business model is for companies to recruit from Kaggle platform. If things like this happen, where the ranking of a competition become rather meaningless, I don't think recruiter will see any value here. As a matter of fact, I found more and more recruiters and people in data science community think Kaggle is just for competition but not real data science experience/projects. I think that is bad for the company as well. \n\nIt takes a long time and effort to build a platform this great and have so many wonderful data scientists (I am relatively new, but already learnt a lot from many discussions posts and kernels). I hope there is a way to keep improving it, keep all the great people around and keep the competition rankings more meaningful. ", "votes": 4 }, { "id": 324233, "postDate": "2018-05-07T13:00:08.803Z", "content": "If people can submit exact same results, even from public kernels, I think it is weird and does not help the ecosystem and should be considered as cunning.
\nhttps://www.kaggle.com/general/48852
Kaggle should allow kernels to be public only after the competition is over.
", "rawMarkdown": "Kaggle should allow kernels to be public only after the competition is over.", "votes": 1 }, { "id": 324721, "postDate": "2018-05-07T22:50:29.727Z", "content": "I disagree with you Siddharth Yadav.\nAllowing kernels to be public ONLY after the competition is way too harsh. It is good for kaggle beginners like me to have some incentive to find solutions while you are entering and workng hard on the competition. Everyone progress. Bombshell kaggle at the last minute are somewhat debatable but at least good explanation and exploratory phases are very helpful to push up your thinking... or at least mine :-)
", "rawMarkdown": "I disagree with you Siddharth Yadav.\nAllowing kernels to be public ONLY after the competition is way too harsh. It is good for kaggle beginners like me to have some incentive to find solutions while you are entering and workng hard on the competition. Everyone progress. Bombshell kaggle at the last minute are somewhat debatable but at least good explanation and exploratory phases are very helpful to push up your thinking... or at least mine :-)" } ] }, { "id": 324894, "postDate": "2018-05-08T01:22:50.803Z", "content": "Was really disappointed to see the blends not be punished for overfitting. I guess leaderboard is kind of a validation set :/
\n\nMy final model: \nPublic AUC .9812 Private AUC .9822\nBlend:\nPublic AUC .9812 Private AUC .9820
", "rawMarkdown": "Was really disappointed to see the blends not be punished for overfitting. I guess leaderboard is kind of a validation set :/\n\nMy final model: \nPublic AUC .9812 Private AUC .9822\nBlend:\nPublic AUC .9812 Private AUC .9820", "votes": 2 }, { "id": 324401, "postDate": "2018-05-07T17:06:48.270Z", "content": "I am new and naive on this subject, but would it be possible or feasible to handle this issue the way patents on IP (intellectual property) are handled? So many products have multiple patent licensing agreements among many patent holders. Obviously if the administrative overhead would be too much, it would not be worth it. Or if things go the way of patent trolls and endless \"litigation\" that's no good either. If there is a way to give credit for an original idea to the originator, that makes sense to me and seems fair.
", "rawMarkdown": "I am new and naive on this subject, but would it be possible or feasible to handle this issue the way patents on IP (intellectual property) are handled? So many products have multiple patent licensing agreements among many patent holders. Obviously if the administrative overhead would be too much, it would not be worth it. Or if things go the way of patent trolls and endless \"litigation\" that's no good either. If there is a way to give credit for an original idea to the originator, that makes sense to me and seems fair. ", "votes": 2 }, { "id": 325136, "postDate": "2018-05-08T06:51:36.480Z", "content": "As a beginner, I think there should be a closed period for games near the end.
", "rawMarkdown": "As a beginner, I think there should be a closed period for games near the end.", "votes": 3 }, { "id": 324245, "postDate": "2018-05-07T13:25:09.623Z", "content": "First of all, this is a great post. I agree with most of the things you mentioned here.
\n\nI will tell you about my experience. I joined like 5 months ago and was unable to approach any competition other than Titanic. For some time, I questioned my decision of joining kaggle. I thought it was too early. That time kernels came to my rescue. I have learned something new while reading kernels every day. I would say there has been a great improvement in my performance as a data scientist. Kernels have surely changed the kaggle ecosystem. Now, people can learn from others' work. There is so much content at one place that you don't have to go anywhere else to increase your knowledge.
\n\nBut on the other side, kernels affect competitions by increasing the ranks of people who just fork and submit. If this has resulted in older users straying away from the platform, it is more serious than I thought. To solve this, I would recommend something like providing proof of work while submitting for competitions. Everyone should also upload their scripts used for the competition. Now, if they prepare kernels on kaggle itself and wish to make them public, they should be able to do that only after the competition is over.
\n\nI don't agree with your suggestion of introducing downvotes for kernels. It takes significant amount of time and hard work to create a kernel. If one can't praise someone's hard work, he/she should also not degrade it by providing downvotes. Downvotes can bring about a negative ecosystem on this platform.
", "rawMarkdown": "First of all, this is a great post. I agree with most of the things you mentioned here. \n\nI will tell you about my experience. I joined like 5 months ago and was unable to approach any competition other than Titanic. For some time, I questioned my decision of joining kaggle. I thought it was too early. That time kernels came to my rescue. I have learned something new while reading kernels every day. I would say there has been a great improvement in my performance as a data scientist. Kernels have surely changed the kaggle ecosystem. Now, people can learn from others' work. There is so much content at one place that you don't have to go anywhere else to increase your knowledge.\n\nBut on the other side, kernels affect competitions by increasing the ranks of people who just fork and submit. If this has resulted in older users straying away from the platform, it is more serious than I thought. To solve this, I would recommend something like providing proof of work while submitting for competitions. Everyone should also upload their scripts used for the competition. Now, if they prepare kernels on kaggle itself and wish to make them public, they should be able to do that only after the competition is over. \n\nI don't agree with your suggestion of introducing downvotes for kernels. It takes significant amount of time and hard work to create a kernel. If one can't praise someone's hard work, he/she should also not degrade it by providing downvotes. Downvotes can bring about a negative ecosystem on this platform. ", "votes": 1 }, { "id": 325414, "postDate": "2018-05-08T11:49:37.673Z", "content": "Maybe we need a few pre-checkings before metric calculation? Like numerai's scoring part.
\n\n\n\n\n...Copied from https://numer.ai/learn
\n \nConsistency measures the percentage of eras in which a model achieves a logloss better than the benchmark. Numerai wants models that work well consistently across eras. Only models with consistency above 58% are considered consistent.
\n \nOriginality is a measure of whether a set of predictions is uncorrelated with predictions already submitted. Numerai wants to encourage new models over duplicate submissions.
\n \nConcordance is a measure of whether predictions on the validation set, test set, and live set appread to be generated by the same model. A data scientist who submits perfect answers on the validation set is unlikely to achieve concordance.
\n
While it may be quite difficult to design perfect checking functions, Originality should be taken care, at least...
", "rawMarkdown": "Maybe we need a few *pre-checkings* before metric calculation? Like [numerai's scoring part][1].\n> ...*Copied from https://numer.ai/learn*\n> \n> **Consistency** measures the percentage of eras in which a model achieves a logloss better than the benchmark. Numerai wants models that work well consistently across eras. Only models with consistency above 58% are considered consistent.\n> \n> **Originality** is a measure of whether a set of predictions is uncorrelated with predictions already submitted. Numerai wants to encourage new models over duplicate submissions.\n> \n> **Concordance** is a measure of whether predictions on the validation set, test set, and live set appread to be generated by the same model. A data scientist who submits perfect answers on the validation set is unlikely to achieve concordance.\n\nWhile it may be quite difficult to design perfect checking functions, **Originality** should be taken care, at least...\n [1]: https://numer.ai/learn", "votes": 1, "replies": [ { "id": 325416, "postDate": "2018-05-08T11:54:20.373Z", "content": "Originality metric in numer.ai isn't used anymore - it is easy to add a small amount of noise to overcome this - just saying
", "rawMarkdown": "Originality metric in numer.ai isn't used anymore - it is easy to add a small amount of noise to overcome this - just saying", "votes": 1 }, { "id": 325425, "postDate": "2018-05-08T12:00:02.633Z", "content": "thanks! didn't know that...
", "rawMarkdown": "thanks! didn't know that..." }, { "id": 325509, "postDate": "2018-05-08T13:40:35.570Z", "content": "\n\n\nit is easy to add a small amount of noise to overcome this -
\n
This may be complex enough to filter out the cut and paste specialists.
", "rawMarkdown": "> it is easy to add a small amount of noise to overcome this - \n\nThis may be complex enough to filter out the cut and paste specialists.", "votes": 5 } ] }, { "id": 324529, "postDate": "2018-05-07T19:16:28.877Z", "content": "I'm new and obviously have no experience with such events. From what I can tell it would be impossible for Kaggle to identify who used \"the kernel\" only by using the submission. Obviously it's easy if its a perfect match but not so easy if any kind of blending has been done.
\n\nOne solution I see for people that deserve medals to get their hard work properly rewarded would be to ask people to submit the model that generates the solution. This way it should be a bit simpler to identify and disqualify users of \"the kernel\". If no model is uploaded the medal should not be awarded.
\n\nWhat do you guys think?
", "rawMarkdown": "I'm new and obviously have no experience with such events. From what I can tell it would be impossible for Kaggle to identify who used \"the kernel\" only by using the submission. Obviously it's easy if its a perfect match but not so easy if any kind of blending has been done. \n\nOne solution I see for people that deserve medals to get their hard work properly rewarded would be to ask people to submit the model that generates the solution. This way it should be a bit simpler to identify and disqualify users of \"the kernel\". If no model is uploaded the medal should not be awarded. \n\nWhat do you guys think?", "votes": 1, "replies": [ { "id": 324540, "postDate": "2018-05-07T19:25:04.403Z", "content": "It is not against the rules to use a publicly available kernel. As much as I don't like it, there is nothing that can be done. Making up rules in the last five hours is a lot worse than someone publishing a high scoring public kernel. Even more so since this was a well known and discussed possibility. Looking at the author previous posts it seems that he wanted to prove a point more than anything else.
", "rawMarkdown": "It is not against the rules to use a publicly available kernel. As much as I don't like it, there is nothing that can be done. Making up rules in the last five hours is a lot worse than someone publishing a high scoring public kernel. Even more so since this was a well known and discussed possibility. Looking at the author previous posts it seems that he wanted to prove a point more than anything else.", "votes": 3 }, { "id": 324545, "postDate": "2018-05-07T19:29:38.260Z", "content": "Got. Well then ... sorry for the guys that lost their medals. Now that makes me wander what is Kaggle looking for when they filter out \"cheaters\" ... Is it only multiple submissions?
", "rawMarkdown": "Got. Well then ... sorry for the guys that lost their medals. Now that makes me wander what is Kaggle looking for when they filter out \"cheaters\" ... Is it only multiple submissions? ", "votes": 2 }, { "id": 324548, "postDate": "2018-05-07T19:31:55.427Z", "content": "Unfortunately there is nothing more we can do by just ignoring this kernel. Good idea is to temporarily close the possibility of uploading kernels ~1 week before deadline.
", "rawMarkdown": "Unfortunately there is nothing more we can do by just ignoring this kernel. Good idea is to temporarily close the possibility of uploading kernels ~1 week before deadline.", "votes": 7 }, { "id": 325148, "postDate": "2018-05-08T07:09:24.650Z", "content": "@Mihai: Kaggle's anti-cheating heuristics are an arms race; they won't disclose them publicly and presumably constantly refine after each competition: I would imagine they look at IP addresses between competitors, identical/near-identical submissions between competitors (perhaps using some hash or signature), unusual late changes in registering, activity, rate of submission, leaderboard jumps, suspicious or non-human-like submission sequence or timing between user A and B, etc. e.g. a user that registers late and does not submit much until the last week, then suddenly jumps, esp. where that submission is similar to others and close in time. Just my informed guesswork, we can't know. Not dissimilar to this competition itself :)
", "rawMarkdown": "@Mihai: Kaggle's anti-cheating heuristics are an arms race; they won't disclose them publicly and presumably constantly refine after each competition: I would imagine they look at IP addresses between competitors, identical/near-identical submissions between competitors (perhaps using some hash or signature), unusual late changes in registering, activity, rate of submission, leaderboard jumps, suspicious or non-human-like submission sequence or timing between user A and B, etc. e.g. a user that registers late and does not submit much until the last week, then suddenly jumps, esp. where that submission is similar to others and close in time. Just my informed guesswork, we can't know. Not dissimilar to this competition itself :)", "votes": 2 } ] }, { "id": 325426, "postDate": "2018-05-08T12:01:10.650Z", "content": "+1
", "rawMarkdown": "+1", "votes": 2 }, { "id": 325578, "postDate": "2018-05-08T15:26:43.573Z", "content": "I consider Kaggle as a educational platform, and the kernels are best materials to learn data science.\nI think it is impossible to ban the kernels in any forms, because it violets the spirit of Kaggle.\nKernels are public resources and every participant share the results, which helps everyone to improve.\nHowever, why not AT LEAST, AT LEAST, AT LEAST blend the results on our own.
\n\nIn short:\nMain problem = Public Blending Kernels
", "rawMarkdown": "I consider Kaggle as a educational platform, and the kernels are best materials to learn data science.\nI think it is impossible to ban the kernels in any forms, because it violets the spirit of Kaggle.\nKernels are public resources and every participant share the results, which helps everyone to improve.\nHowever, why not AT LEAST, AT LEAST, AT LEAST blend the results on our own.\n\nIn short:\nMain problem = Public Blending Kernels" }, { "id": 324458, "postDate": "2018-05-07T18:01:26.300Z", "content": "Generally, I find high-scoring public LB models to actually be pretty helpful as I can blend them with my own and improve my own score. Now, I totally get the frustration involved in the hundreds of blending posts - they should be used to serve a purpose such as demonstrate that highly diverse low-scoring models can be blended to perform far better, or how you can determine weights for your ensembling model.
", "rawMarkdown": "Generally, I find high-scoring public LB models to actually be pretty helpful as I can blend them with my own and improve my own score. Now, I totally get the frustration involved in the hundreds of blending posts - they should be used to serve a purpose such as demonstrate that highly diverse low-scoring models can be blended to perform far better, or how you can determine weights for your ensembling model.", "votes": 1, "replies": [ { "id": 324463, "postDate": "2018-05-07T18:09:38.557Z", "content": "I try to produce kernels that are not LB killers but can generally be used in future competitions and real life. -- Needless to say they score very poorly with respect to votes and so I am seriously considering not bothering.
\n\nIf you want to get loads of votes just do a super blend or knock up an eda in 30 minutes at the start of a competition.
", "rawMarkdown": "I try to produce kernels that are not LB killers but can generally be used in future competitions and real life. -- Needless to say they score very poorly with respect to votes and so I am seriously considering not bothering.\n\n***If you want to get loads of votes just do a super blend or knock up an eda in 30 minutes at the start of a competition.***", "votes": 10 }, { "id": 324469, "postDate": "2018-05-07T18:18:04.077Z", "content": "I was misinformed by my own self and forgot the competition ends tonight. I would say that he should have released this kernel at least a week earlier, preferably with an explanation or justification. I would feel completely jobbed if I lost my bronze / silver medal after several months of hardwork only because someone released a very high-scoring kernel 8 hours before the competition ended and I wasn't there to get the data.
\n\nI've thought about this for myself. I've been running a submission with LGB-XGB blend for about a week now, and when I get home from school today I'll submit it, and hypothetically it could be bronze-material if only this kernel hadn't been released. (I say it's bronze material because my validation scores have been following the leaderboard very well thus far (r=0.92) and if that pattern were to continue, it would be bronze material excluding these new top-150 solutions.
", "rawMarkdown": "I was misinformed by my own self and forgot the competition ends tonight. I would say that he should have released this kernel at least a week earlier, preferably with an explanation or justification. **I would feel completely jobbed if I lost my bronze / silver medal after several months of hardwork only because someone released a very high-scoring kernel 8 hours before the competition ended and I wasn't there to get the data.**\n\nI've thought about this for myself. I've been running a submission with LGB-XGB blend for about a week now, and when I get home from school today I'll submit it, and hypothetically it could be bronze-material if only this kernel hadn't been released. (I say it's bronze material because my validation scores have been following the leaderboard very well thus far (r=0.92) and if that pattern were to continue, it would be bronze material excluding these new top-150 solutions.", "votes": 3 }, { "id": 324485, "postDate": "2018-05-07T18:31:18.660Z", "content": "A couple of things: first, I don't actually mind that much, although it does bum me out. Second, the OP dropped rank from 153 to 160 to 167 in the past 30 minutes due to these ridiculous submissions.
", "rawMarkdown": "A couple of things: first, I don't actually mind that much, although it does bum me out. Second, the OP dropped rank from 153 to 160 to 167 in the past 30 minutes due to these ridiculous submissions.", "votes": 1 }, { "id": 324624, "postDate": "2018-05-07T21:31:23.193Z", "content": "He has dropped at least 135 ranks in 3.5 hours due to a kernel people keep cheating off of. That seems very unfair.
", "rawMarkdown": "He has dropped at least 135 ranks in 3.5 hours due to a kernel people keep cheating off of. That seems very unfair.", "votes": 2 }, { "id": 324684, "postDate": "2018-05-07T22:17:49.543Z", "content": "Update: 182 ranks in 4 hours.
", "rawMarkdown": "Update: 182 ranks in 4 hours." }, { "id": 325140, "postDate": "2018-05-08T06:58:44.020Z", "content": "\n\n\nMatthew Anderson wrote
\n \n\n\nUpdate: 182 ranks in 4 hours.
\n
But did those other people use the cheating kernel to poison the public LB by getting a high public LB score but not selecting it for their private final score?
", "rawMarkdown": "> **Matthew Anderson wrote**\n> \n> > Update: 182 ranks in 4 hours.\n\nBut did those other people use the cheating kernel to poison the public LB by getting a high public LB score but not selecting it for their private final score? " }, { "id": 325174, "postDate": "2018-05-08T07:31:15.853Z", "content": "All that blending is annoying. I would be much more tempted to enter a competition where the best single model wins (as the added complexity is not worth the extra 0.000001% in real life anyway). Why not just do a comp like this, with proof in private kernels (submit from private kernel only)?
", "rawMarkdown": "All that blending is annoying. I would be much more tempted to enter a competition where the best single model wins (as the added complexity is not worth the extra 0.000001% in real life anyway). Why not just do a comp like this, with proof in private kernels (submit from private kernel only)?", "votes": 1 }, { "id": 325261, "postDate": "2018-05-08T09:04:14.973Z", "content": "@Scirpus, your kernels are great examples of what sharing could be. Thanks for that.
", "rawMarkdown": "@Scirpus, your kernels are great examples of what sharing could be. Thanks for that.", "votes": 1 }, { "id": 325271, "postDate": "2018-05-08T09:14:01.637Z", "content": "Thanks for the encouragement - it means a lot ;)
", "rawMarkdown": "Thanks for the encouragement - it means a lot ;)", "votes": 2 } ] }, { "id": 324771, "postDate": "2018-05-07T23:42:01.020Z", "content": "I understand that dropping 100 positions within a day or within a few hours can be really frustrating. But does it really matter in the end? If you finish #200 or #300? If you get bronze or not? In the beginning it is only the learning process that matters. I personally learned a lot from kernels and improved my skills ever since I started kaggeling. I am very thankful that Kaggle is providing this great platform. So no need to complain, Kaggle is giving their best, especially for starters.
\n\nPS: And if you care about the rank, take the public kernel, blend it with your own model and there you have a jump on LB. That's what I did in my first competition and it worked pretty well. Of course I don't like to drop 100 positions as well :)
", "rawMarkdown": "I understand that dropping 100 positions within a day or within a few hours can be really frustrating. But does it really matter in the end? If you finish #200 or #300? If you get bronze or not? In the beginning it is only the learning process that matters. I personally learned a lot from kernels and improved my skills ever since I started kaggeling. I am very thankful that Kaggle is providing this great platform. So no need to complain, Kaggle is giving their best, especially for starters.\n\nPS: And if you care about the rank, take the public kernel, blend it with your own model and there you have a jump on LB. That's what I did in my first competition and it worked pretty well. Of course I don't like to drop 100 positions as well :)", "votes": -6, "replies": [ { "id": 324774, "postDate": "2018-05-07T23:46:37.863Z", "content": "I dropped 327 positions!
\n\nThis is about learning: yes!! it is about what you can produce not about what you can copy!!
", "rawMarkdown": "I dropped 327 positions!\n\nThis is about learning: yes!! it is about what you can produce not about what you can copy!!", "votes": 4 }, { "id": 324777, "postDate": "2018-05-07T23:51:00.403Z", "content": "Well sometimes you learn from copying too. Play dirty, take the kernel and try to climb up the LB again. This is not against the rules. Everybody wants to win ;)
", "rawMarkdown": "Well sometimes you learn from copying too. Play dirty, take the kernel and try to climb up the LB again. This is not against the rules. Everybody wants to win ;)", "votes": -10 }, { "id": 324811, "postDate": "2018-05-08T00:17:07.967Z", "rawMarkdown": "", "isDeleted": true }, { "id": 324958, "postDate": "2018-05-08T02:15:41.787Z", "content": "I think the reason you saying this is you had pretty high rank and have already had several medals. Say you did not have any medals and were eager winning one. You worked hard climbing to #100 and at the last day, your dropped 300 because of one kernel. I don't think you can learn anything from here. Learn how to blend or learn doing nothing but wait?\nKernels are great resource to learn, I agree, but I don't agree with what you said \"if you care about the rank, take the public kernel, blend it to get higher rank\". It's just not that simple. Well if everybody start blending at the final point, how could you assure your blender is better ? That would be more like a lottery game.\nI don't think you would say same thing if 4th player sharing everything and you became the 400th. Let's say it repeats several times would you still be that optimistic?
", "rawMarkdown": "I think the reason you saying this is you had pretty high rank and have already had several medals. Say you did not have any medals and were eager winning one. You worked hard climbing to #100 and at the last day, your dropped 300 because of one kernel. I don't think you can learn anything from here. Learn how to blend or learn doing nothing but wait?\nKernels are great resource to learn, I agree, but I don't agree with what you said \"if you care about the rank, take the public kernel, blend it to get higher rank\". It's just not that simple. Well if everybody start blending at the final point, how could you assure your blender is better ? That would be more like a lottery game.\nI don't think you would say same thing if 4th player sharing everything and you became the 400th. Let's say it repeats several times would you still be that optimistic? ", "votes": 8 }, { "id": 324982, "postDate": "2018-05-08T02:38:14.243Z", "content": "Ok, i see. So people who what to win something just need to wait for the last day, download some submissions, do blending, think nothing, grab a coffee, win medals, thank you so much.
", "rawMarkdown": "Ok, i see. So people who what to win something **just** need to wait for the last day, download some submissions, do blending, think nothing, grab a coffee, win medals, thank you so much.", "votes": 2 }, { "id": 325201, "postDate": "2018-05-08T08:10:18.617Z", "content": "Don't you think that I also dropped hundreds of positions within days because of kernels in the beginning. I took overfitting kernels and dropped many positions on private LB. But I learned which kernels I can trust and which not. Sometimes it's lottery but often it is not. Nobody assures you that your blend is better but with time you develop intuition for good and bad. You can select two submissions, so take a risky and take a save. After some time you'll have enough experience and public kernels won't get close to your own model...
\n\nOh I'd have been very happy if for example Ahmet would have shared this information a bit earlier. Maybe 12 hours before competition end... it would have been enough to retrain my model :D But he decided to keep it for himself, I don't understand why ;)
", "rawMarkdown": "Don't you think that I also dropped hundreds of positions within days because of kernels in the beginning. I took overfitting kernels and dropped many positions on private LB. But I learned which kernels I can trust and which not. Sometimes it's lottery but often it is not. Nobody assures you that your blend is better but with time you develop intuition for good and bad. You can select two submissions, so take a risky and take a save. After some time you'll have enough experience and public kernels won't get close to your own model...\n\nOh I'd have been very happy if for example [Ahmet][1] would have shared [this information][2] a bit earlier. Maybe 12 hours before competition end... it would have been enough to retrain my model :D But he decided to keep it for himself, I don't understand why ;)\n\n [1]: https://www.kaggle.com/divrikwicky\n [2]: https://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56268", "votes": -2 }, { "id": 325231, "postDate": "2018-05-08T08:42:53.237Z", "content": "In fact, I just realized that the same information was shared a few days ago. But we did not took advantage of it while others did. Too bad...
", "rawMarkdown": "In fact, I just realized that the [same information][1] was shared a few days ago. But we did not took advantage of it while others did. Too bad...\n\n [1]: https://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/55677", "votes": -2 }, { "id": 325494, "postDate": "2018-05-08T13:21:42.913Z", "content": "I don't think you get the point. \"Don't you think that I also dropped hundreds of positions within days because of kernels in the beginning, I took overfitting kernels and dropped many positions on private LB\", People won't care in the early stage drop how many positions, cause they can still learn from it and move on. What can you do in the last second????
", "rawMarkdown": "I don't think you get the point. \"Don't you think that I also dropped hundreds of positions within days because of kernels in the beginning, I took overfitting kernels and dropped many positions on private LB\", People won't care in the early stage drop how many positions, cause they can still learn from it and move on. What can you do in the last second????", "votes": 3 }, { "id": 325512, "postDate": "2018-05-08T13:42:45.013Z", "content": "@YIANG Totally agree with you. At the beginning of competition, who cares your position?? On the contrary, last few hours public high score kernel is certainly nonsense cheating and injustice to people paying really huge effort.
", "rawMarkdown": "@YIANG Totally agree with you. At the beginning of competition, who cares your position?? On the contrary, last few hours public high score kernel is certainly nonsense **cheating** and injustice to people paying really huge effort. ", "votes": 5 } ] }, { "id": 324760, "postDate": "2018-05-07T23:29:33.723Z", "content": "This kind of kernels can be very frustrating, but I would argue that ranks, medals and points don't have much value themselves. It is knowledge behind them that matters.
", "rawMarkdown": "This kind of kernels can be very frustrating, but I would argue that ranks, medals and points don't have much value themselves. It is knowledge behind them that matters.", "votes": -4, "replies": [ { "id": 325391, "postDate": "2018-05-08T11:14:33.300Z", "content": "I don't get it why saying that \"ranks, medals and points don't have much value themselves. It is knowledge behind them that matters.\" is being downvoted. For me it looks like a good description of what Kaggle is. Tragic would be if people stopped sharing knowledge.
", "rawMarkdown": "I don't get it why saying that *\"ranks, medals and points don't have much value themselves. It is knowledge behind them that matters.\"* is being downvoted. For me it looks like a good description of what Kaggle is. Tragic would be if people stopped sharing knowledge. " }, { "id": 325401, "postDate": "2018-05-08T11:35:48.653Z", "content": "I didn't downvote, but I believe medals do have some value for newcomers. It's a resume padder, showing that you can apply your knowledge to some problems and do well.
\n\nOf course, the knowledge itself is more important, but saying medals don't have much value (for newcomers) is wrong in my opinion.
", "rawMarkdown": "I didn't downvote, but I believe medals do have some value for newcomers. It's a resume padder, showing that you can apply your knowledge to some problems and do well.\n\nOf course, the knowledge itself is more important, but saying medals don't have much value (for newcomers) is wrong in my opinion.", "votes": 5 } ] }, { "id": 327851, "postDate": "2018-05-12T16:53:56.623Z", "content": "How about we keep the kernel private and adding the share option for private kernel, that sound ridiculous but that might keep the reputation.
", "rawMarkdown": "How about we keep the kernel private and adding the share option for private kernel, that sound ridiculous but that might keep the reputation. " }, { "id": 325660, "postDate": "2018-05-08T17:14:58.163Z", "content": "Kaggle is probably doing something for it.. who knows? \nThey are yet to finalise the final standing..
", "rawMarkdown": "Kaggle is probably doing something for it.. who knows? \nThey are yet to finalise the final standing.." }, { "id": 324490, "postDate": "2018-05-07T18:41:37.507Z", "content": "It's a pity seeing that competitors like us are using that infamous kernel.
", "rawMarkdown": "It's a pity seeing that competitors like us are using that infamous kernel." }, { "id": 343430, "postDate": "2018-06-15T09:58:40.437Z", "rawMarkdown": "", "votes": -2, "isDeleted": true }, { "id": 326171, "postDate": "2018-05-09T11:48:09.450Z", "rawMarkdown": "", "isDeleted": true }, { "id": 325574, "postDate": "2018-05-08T15:11:36.023Z", "rawMarkdown": "", "votes": -9, "isDeleted": true, "replies": [ { "id": 325585, "postDate": "2018-05-08T15:36:29.180Z", "content": "I knew about the kernel, I downloaded the the data, I looked at it, I had a spare submission. I had two submissions to select but really one single model up to then, so your second point doesn't apply.
\n\nI decided the bronze (or silver) is not worth my integrity.
\n\nWhere do I fit in your universe?
\n\nEdit: BTW, since I haven't used my submission for the first CSV, that means it was still available for the blended CSV too.
", "rawMarkdown": "I knew about the kernel, I downloaded the the data, I looked at it, I had a spare submission. I had two submissions to select but really one single model up to then, so your second point doesn't apply.\n\nI decided the bronze (or silver) is not worth my integrity.\n\nWhere do I fit in your universe?\n\nEdit: BTW, since I haven't used my submission for the first CSV, that means it was still available for the blended CSV too.", "votes": 3 }, { "id": 325597, "postDate": "2018-05-08T15:56:44.010Z", "rawMarkdown": "", "votes": -1, "isDeleted": true } ] }, { "id": 325451, "postDate": "2018-05-08T12:31:42.117Z", "rawMarkdown": "", "isDeleted": true }, { "id": 324267, "postDate": "2018-05-07T14:00:31.583Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 324285, "postDate": "2018-05-07T14:25:01.680Z", "content": "Well, @giba managed to be top Kaggler for years while having a job. There is hope therefore ;)
", "rawMarkdown": "Well, @giba managed to be top Kaggler for years while having a job. There is hope therefore ;)", "votes": 1 }, { "id": 324311, "postDate": "2018-05-07T15:03:12.877Z", "content": "IME - the job is the easier of the two obligations to manage. It's the wife + kid(s) that are super anti-Kaggle :-)
", "rawMarkdown": "IME - the job is the easier of the two obligations to manage. It's the wife + kid(s) that are super anti-Kaggle :-)", "votes": 7 }, { "id": 324687, "postDate": "2018-05-07T22:19:04.657Z", "content": "Good to know that @giba managed so well. It is an exemple for us!
", "rawMarkdown": "Good to know that @giba managed so well. It is an exemple for us!", "votes": -1 } ] } ], "comments": [ { "id": 324898, "author_name": "Hiroshi Arai", "author_url": "", "post_date": "2018-05-08T01:25:41.393000", "content": "I know what we learned is much more important than the score, but many companies are using Kaggle as one of evaluation metrics for hiring data scientists. Kaggle itself has the job board, and many companies are posting jobs there.
\n\nI think Kaggle should do something to keep the reputation unless these companies are seeking \"Blending Scientists\".
", "votes": 56, "replies": [ { "id": 325254, "author_name": "CPMP", "author_url": "", "post_date": "2018-05-08T09:01:39.263000", "content": "Best comment I read on this issue!
", "votes": 6, "replies": [] }, { "id": 325301, "author_name": "Erik Bruin", "author_url": "", "post_date": "2018-05-08T09:36:35.117000", "content": "I may have a bit of a simplistic mind, but the solution seems so simple to me.\n1. Only accept best single models (indeed, companies are not seeking 'Blending Scientists')\n2. Models need to be run through private kernels\n3. The score of this run is the submission (no separate csv submissions)\n4. The private kernels could possibly all be made public automatically after comp ends
", "votes": 5, "replies": [] }, { "id": 325328, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-08T10:08:34.230000", "content": "I think you are throwing the baby with the bath water - it would work but we would be restricted to toy models essentially - Kaggle does run certain competitions like that but I hate them as I find they are too restrictive
", "votes": 1, "replies": [] }, { "id": 325355, "author_name": "Erik Bruin", "author_url": "", "post_date": "2018-05-08T10:38:23.033000", "content": "Hmm....ok, I see. But isn't toying models exactly what companies are looking for? I think that Netflix solution also never got implemented. I prefer to learn only things that are useful in real life. But anyway, that probably just means that I have to wait for such a competition (and do kernels in the meantime) ;-).
", "votes": 1, "replies": [] } ] }, { "id": 324378, "author_name": "Ahmet Erdem", "author_url": "", "post_date": "2018-05-07T16:38:19.373000", "content": "Kaggle should disable kernel sharing during the last week of the competition. I think this is the most optimal solution.
", "votes": 58, "replies": [ { "id": 324430, "author_name": "Ų¹Ų«Ł Ų§Ł", "author_url": "", "post_date": "2018-05-07T17:30:45.807000", "content": "This honestly seems like the most straightforward solution to the problem. Downvoting kernels won't really fix the issue (though I agree it's a nice to have feature). And coming up with a filtering mechanism is a game of cat-and-mouse, since noise can be added to the output. Better just to shut it down once at the same time that team merge deadline occurs. Alternatively, use something like Kaggle Kernel-Ranking, where users have to hit Kernel-Expert, etc. before they can open new Kernels in the last week.
", "votes": 6, "replies": [] }, { "id": 324470, "author_name": "alijs", "author_url": "", "post_date": "2018-05-07T18:18:58.317000", "content": "+1\nI would also vote for disabling public kernels in last competition week. Of course that probably would not prevent someone hunting for cheap up-votes from e.g. posting links to their Githubs in discussions, but I think that would be a step into the right direction. And discussions at least can be downvoted, if high scoring solution is posted in the last day of the competition ;)
", "votes": 8, "replies": [] }, { "id": 324473, "author_name": "Matthew Anderson", "author_url": "", "post_date": "2018-05-07T18:20:52.810000", "content": "My concern about blocking kernels the last week would be the same as yours - Github repos. I think posting a high scoring solution on that would be even more unfair because its not as transparent and clear as the Kaggle kernel ecosystem.
", "votes": 3, "replies": [] }, { "id": 324478, "author_name": "Ų¹Ų«Ł Ų§Ł", "author_url": "", "post_date": "2018-05-07T18:25:52.907000", "content": "I'm not worried about github repos. For one, that would be private sharing so they (should) end up disqualified. I recall in a past competition, some people were having a discussion on KaggleNoobs slack and it started getting to code sharing were told they had bring it onto the forums---which they did---so I don't imagine github being much different.
\n\nMoreover, one has to look at the intent behind these perpetrators. Their goal is to either get a lot of upvotes, or to destroy the integrity of the competition. In the former case, disabling kernels w/ team merger deadline solves it. In the later case, it makes it a LOT harder to broadcast results to the masses, so net-net win imo.
", "votes": 3, "replies": [] }, { "id": 324479, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-07T18:26:22.207000", "content": "Just disable links in the last week in the discussion pages too - at least you can flag discussion posts
", "votes": 2, "replies": [] }, { "id": 324606, "author_name": "yulia", "author_url": "", "post_date": "2018-05-07T21:08:40.027000", "content": "@alijs But these would be clear rule violations and threat of disqualification should prevent them
", "votes": 1, "replies": [] }, { "id": 324623, "author_name": "Daniel J Brooks", "author_url": "", "post_date": "2018-05-07T21:30:37.943000", "content": "This is clearly the best solution. It allows for shared context, but not shared solutions.
", "votes": 0, "replies": [] }, { "id": 1600274, "author_name": "Pascal Pfeiffer", "author_url": "", "post_date": "2021-11-30T08:48:55.357000", "content": "This aged well
", "votes": 0, "replies": [] } ] }, { "id": 325437, "author_name": "Wenjie Bai", "author_url": "", "post_date": "2018-05-08T12:16:21.380000", "content": "Anyone help me debug this? It looks quiet runnable for kaggleš
\n\nimport numpy as np\nimport warnings\nfrom pandas.tseries.offsets import DateOffset\n\ndef win_medals():\n '''\n Only works to silver and bronze, have fun!\n '''\n if Competition == 'kaggle':\n assert Remaning_hours_to_deadline <= DateOffset(hours=12), \"Too early man,it's not time!\"\n\n grab_a_coffee()\n click_kernels_button()\n csvs = download_highest_score_csv(n=10)\n submits = do_blending(csvs, random_state=np.random.randint(1,100))\n submit_predictions(submits)\n\n if good_luck:\n return SILVER\n else:\n return BRONZE\n else:\n pass\n\nif __name__ == 'main':\n medals = win_medals()\n if entire_running_time >= DateOffset(hours=4)\n warnings.warn('You are wasting too much time! You should be quicker, try next time use only brute force without brain!')\nHAHA!
", "votes": 5, "replies": [] }, { "id": 325532, "author_name": "Pranav Pandya", "author_url": "", "post_date": "2018-05-08T14:08:16.997000", "content": ":) good one @Wenjie
", "votes": 1, "replies": [] }, { "id": 325535, "author_name": "Sun ZhiHao", "author_url": "", "post_date": "2018-05-08T14:10:03.393000", "content": "Why are you so xiu~
", "votes": 1, "replies": [] }, { "id": 325537, "author_name": "Richard Marto", "author_url": "", "post_date": "2018-05-08T14:14:16.517000", "content": "can I use this code at work or only at kaggle ?
", "votes": 2, "replies": [] }, { "id": 325560, "author_name": "Araks Stepanyan", "author_url": "", "post_date": "2018-05-08T14:48:12.913000", "content": "This was awesome, can't stop laughing! The best thing that happened after the yesterday's deadline :) Thanks...
", "votes": 2, "replies": [] }, { "id": 325587, "author_name": "Wenjie Bai", "author_url": "", "post_date": "2018-05-08T15:45:59.933000", "content": "Hahaha~ Thank you guys for all these comments!~ Life is full of hopes and joys, right? I believe we who really think independently and work hard earn most whether for short term or long term. Hope to meet you guys in the following competition or maybe form a team to enjoy the learning process/competition together!š
", "votes": 7, "replies": [] }, { "id": 325848, "author_name": "Masser", "author_url": "", "post_date": "2018-05-09T00:46:32.207000", "content": "Cool! I want to use this for next competition!, but I hope Kaggle conducts competitions appropriately in the future.
", "votes": 2, "replies": [] }, { "id": 325864, "author_name": "YulinGUO", "author_url": "", "post_date": "2018-05-09T01:27:59.230000", "content": "So pi~~~
", "votes": 1, "replies": [] }, { "id": 326100, "author_name": "Nathan Lauga", "author_url": "", "post_date": "2018-05-09T09:27:46.873000", "content": "Great code ! Do you have the same in R ? ;)
", "votes": 2, "replies": [] }, { "id": 326186, "author_name": "Malte Nalenz", "author_url": "", "post_date": "2018-05-09T12:29:53.887000", "content": "Cant find the submit button, please help!
", "votes": 1, "replies": [] }, { "id": 326220, "author_name": "Gerryl", "author_url": "", "post_date": "2018-05-09T13:08:55.150000", "content": "ęę¶ēēē®
", "votes": 0, "replies": [] }, { "id": 326252, "author_name": "Rayarrow", "author_url": "", "post_date": "2018-05-09T13:38:52.087000", "content": "You're definitely not äŗŗé é©, You're ēēē®.
", "votes": 0, "replies": [] } ] }, { "id": 324664, "author_name": "ĪĪ±ĻĪ¹ĪæĻ ĪĪ¹ĻĪ±Ī·Ī»Ī¹Ī“Ī·Ļ KazAnova", "author_url": "", "post_date": "2018-05-07T22:03:30.553000", "content": "Many people have suggested to block kernels in the last week. We have been saying this for ages. Why kaggle hasn't acted on this?
\n\nI am curious on the reasoning that kernels are still allowed in the last week ... I have honestly not seen much opposition for this suggestion (if any).
\n\nOn the contrary I think kaggle seems to want this. Kernels and discussions are gamified and you can get points/ranks from these elements . If someone is not very high in competitions or he/she targets to get master/grandmaster through kernels, he/she can unleash a high scoring kernel near the end to get points/votes.
", "votes": 42, "replies": [ { "id": 324683, "author_name": "Eric", "author_url": "", "post_date": "2018-05-07T22:17:47.073000", "content": "I believe highly ranked kernel publication at the last minute gives to the game more stamina. It is like throwing some oil on the fire. It fires up more and makes the game more popular... It reminds me ebay auction where people wait until the last minute to make their bid...
", "votes": -11, "replies": [] }, { "id": 324685, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-07T22:18:26.917000", "content": "The only reason I can think of is that it inflates the scores which looks better for kaggle. I have wracked my brains for a less cynical explanation but I cannot find one
", "votes": 2, "replies": [] }, { "id": 324689, "author_name": "Joe Eddy", "author_url": "", "post_date": "2018-05-07T22:22:05.457000", "content": "\n\n\neric wrote
\n \n\n\nI believe highly ranked kernel publication at the last minute gives to the game more stamina. It is like throwing some oil on the fire. It fires up more and makes the game more popular... It reminds me ebay auction where people wait until the last minute to make their bid...
\n
I don't think that several month long data science competitions that involve 200 million datapoints should be reminiscent of last minute ebay auctions.
", "votes": 16, "replies": [] }, { "id": 324692, "author_name": "Izmaylov Konstantin", "author_url": "", "post_date": "2018-05-07T22:23:55.640000", "content": "Hope that this kernel is an overfitted joke and it will be the epic competition.
", "votes": 4, "replies": [] }, { "id": 324723, "author_name": "Eric", "author_url": "", "post_date": "2018-05-07T22:55:32.507000", "content": "Scirpus, agree that it inflates scores and makes the kaggle community look smarter!
", "votes": -6, "replies": [] }, { "id": 325132, "author_name": "Aljaž", "author_url": "", "post_date": "2018-05-08T06:44:51.960000", "content": "Kernels are one part of the problem. What about the discussions? If I remember correctly, a huge spoiler in the Instacart Market Basket competition was shared via github. In the end I finished some ~ 150 places lower, and lost most of the motivation after more than a month of active participation.
", "votes": 1, "replies": [] }, { "id": 325316, "author_name": "Myles O'Neill", "author_url": "", "post_date": "2018-05-08T09:49:26.193000", "content": "As @inversion mentioned in this thread this is a very serious issue that we are taking very seriously within Kaggle and have been thinking about for a long time, obviously.
\n\nJust to add a little bit of context here. For about a year we used to disable kernel submission to competitions during the last week of a competition. Unfortunately this didn't stop people sharing the code on kernels anyway (or in discussion posts) and it caused a lot of users to write to us upset because they didn't understand why submission wasn't working.
\n\nIn June last year we added private kernels to the site as the default, this has opened up some more possibilities that we are going to be actively talking about. But its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.
", "votes": 11, "replies": [] }, { "id": 325333, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-08T10:12:22.783000", "content": "@ myles. Thanks for the writing in this forum.
\n\nI have a few specific questions.
\n\nWhat has happened in literally the last few hours is that hundreds of users have uploaded the same file as their input data file. Does Kaggle consider this as a violation of rules? And by violation I mean something that is serious enough for removal from the competition.
\n\nIf a kernel was public and everyone ran that same kernel for a similar score there is nothing anyone can do about it. But here, people have directly uploaded the file as their own private data set. Isn't the nature of this problem a bit more grave and shouldn't this be penalized?
\n\nPlease do let us Kagglers know where Kaggle stands on this because if this is not a violation perhaps in the next competition when someone leaks a kernel I'll be sure to blend it myself - especially since I wouldn't run the risk of disqualification.
\n\nThe thought of safeguarding my position did occur to me yesterday but I DID NOT USE the file only, only because I was SO SURE that Kaggle would consider this a violation.
\n\nRegards\nShanth
", "votes": 2, "replies": [] }, { "id": 325394, "author_name": "ĪĪ±ĻĪ¹ĪæĻ ĪĪ¹ĻĪ±Ī·Ī»Ī¹Ī“Ī·Ļ KazAnova", "author_url": "", "post_date": "2018-05-08T11:22:00.467000", "content": "hey @Myles O'Neill
\n\nThank you for the response - I am glad kaggle is looking into this. I agree that if someone is determined to pass on the information , he/she will find a way to do it - but it is a whole different story when you essentially encourage it ( with potential kernel upvotes).
\n\nIt is a culture thing too ... People should somehow be aware that this is not how things should work , even if it is not strictly on the rules. I dont think the culture is there now.
", "votes": 8, "replies": [] }, { "id": 325400, "author_name": "Meyk", "author_url": "", "post_date": "2018-05-08T11:33:16.490000", "content": "\n\n\nMyles O'Neill wrote
\n \n\n\n \nAs @inversion mentioned in this thread this is a very serious issue that we are taking very seriously within Kaggle and have been thinking about for a long time, obviously.
\nJust to add a little bit of context here. For about a year we used to disable kernel submission to competitions during the last week of a competition. Unfortunately this didn't stop people sharing the code on kernels anyway (or in discussion posts) and it caused a lot of users to write to us upset because they didn't understand why submission wasn't working.
\n \nIn June last year we added private kernels to the site as the default, this has opened up some more possibilities that we are going to be actively talking about. But its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.
\n
It exists solution that may works. If some people are not mature enough to understand what a 'fair competition' is then it have to be imposed. Like in that case - disabling kernels is not enough because of discussion sharing possibility. Thus, I think, it should be forbidden to share your kernels / solutions / csv/ etc. in any way during last week of competition. It is very strict, but efficient.
\n\nI wish that everyone can understand 'fair play' but this is the ideal world - in which we are not living :(
\n\nThanks for your involvement in that case!
", "votes": 0, "replies": [] }, { "id": 325404, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-08T11:39:01.923000", "content": "Shutdown Discussion in the last week too - draconian - probably - but it just might work
", "votes": 2, "replies": [] }, { "id": 325423, "author_name": "Meyk", "author_url": "", "post_date": "2018-05-08T11:59:55.383000", "content": "I would like to avoid this method, but you have a point ;)
", "votes": 1, "replies": [] }, { "id": 325482, "author_name": "Pranav Pandya", "author_url": "", "post_date": "2018-05-08T13:06:57.187000", "content": "\n\n\nMyles O'Neill wrote
\n \nBut its important to realize that if someone is trying to share their code they can always go around anything we do by just posting it on github and sharing a link. Blocking kernels can't actually solve this problem on its own, it only stops someone sharing unintentionally.
\n
I believe there a huge gap between users checking kernels and users reading posts on discussion forum. So blocking kernels can certainly reduce the major impact. I also think it's a good idea let the community decide what to do with such kernels or links posted on last day if Kaggle can't take any solid action on LB destroyers.
\n\nFor example, allowing specific user group to delete the kernel or link posted on last day . This user group can be something like participants with contributor or above badge and having atleast 50 submissions.
\n\nLet's say if 20 users from this special user group indicates that kernel/link is harmful to competition, they can simply take it down.
", "votes": 2, "replies": [] }, { "id": 325915, "author_name": "MitchelFung", "author_url": "", "post_date": "2018-05-09T04:02:16.533000", "content": "I think a combination between disabling kernels in the last week and banning users who post/share kernels in discussions in the last week should solve the issue. If people try to get around the rules/system then they should be banned, similar to users who create multiple accounts to get extra submissions. Only issue that would leave is people with public githubs... but at that point I feel like it would be more that they were just careless on keeping their github private than maliciously sharing code.
", "votes": 0, "replies": [] }, { "id": 326388, "author_name": "Araks Stepanyan", "author_url": "", "post_date": "2018-05-09T16:16:50.380000", "content": "Check this out, the leaderboard progression animation: https://www.kaggle.com/inversion/talkingdata-leaderboard-progression/code
", "votes": 0, "replies": [] } ] }, { "id": 324476, "author_name": "inversion", "author_url": "", "post_date": "2018-05-07T18:24:02.497000", "content": "This is a discussion point that gets re-ignited every few months, and becomes more challenging as Kaggle continues to add more functionality to Kernels and Datasets.
\n\nWe are following this closely, and will be having focused discussions internally on how we can both (a) continue to provide value-added tools to the data science community as well as (b) minimize the amount of frustration that might arise during competitions because of these tools.
\n\nPlease continue to post your thoughts and suggests in this thread. Your frank feedback is very much appreciated.
", "votes": 37, "replies": [ { "id": 324504, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-07T18:51:25.820000", "content": "@ Inversion
\n\nWe exchanged a couple of messages a few days ago where you gave me confirmation that CSV files can be used for submission directly as long as they are NOT shared outside the team and are maintained as PRIVATE data sets. \n CSV CANNOT BE SHARED
\n\nUnfortunately, someone chose to IGNORE that rule. In fact, I shared this above link with the person who had posted the kernel and within a few minutes the kernel was taken down.
\n\nCan the Kaggle Team do something about this Please?
\n\nIf someone published a kernel and everyone ran it to get to this score, I guess there is nothing Kaggle can do about it. But , in this case the CSV file was simply downloaded and then uploaded again to get to an output.
\n\nThe csv that produces the 0.9811 score with the EXACT SAME OUTPUT would have been by multiple people. Is this not a violation of competition rules? Again, running a public kernel for a high score is not in the spirit of the competition either but what has happened this time seems to be like an outright violation.
\n\nWould it be fair to disqualify anyone having the 0.9811 file in their Data files? Will the Kaggle team be able to do that in the evaluation process? If this process may take time, those of us who worked hard for more than a month will be more than happy to wait for a few more days till the final ranks come out.
\n\nI very sincerely request the Kaggle team to take this into consideration while evaluating the final competition results.
\n\nI have learnt much from Kaggle and will continue to do so. Me and a lot of others will hope for a favourable resolution on this issue. I will accept any final decision that Kaggle makes.
\n\nRegards\nShanth
", "votes": 6, "replies": [] }, { "id": 324505, "author_name": "Shitian Ni", "author_url": "", "post_date": "2018-05-07T18:52:56.783000", "content": "@inversion Thank you for your kindness. I suggest you delete copy & paste, download & submit from leaderboard. It is simple to detect, solve the most frustrating part which is people who does nothing get better ranks. Same repeated result on leaderboard doesnāt benefit anyone except lazy cheaters.
", "votes": 4, "replies": [] }, { "id": 324509, "author_name": "Ų¹Ų«Ł Ų§Ł", "author_url": "", "post_date": "2018-05-07T18:57:20.717000", "content": "Those suggestions are good @shanth84
\n\nBut rather than disqualifying them, just disqualify that entry alone. A lot of hard workers who got to 0.9800 or wherever they got to succumbed to the infamous \"9811\" kernel and submitted that too, along with their own predictions out of frustration. I don't think those ppl should be DQ'd. Just that submission.
", "votes": 6, "replies": [] }, { "id": 324513, "author_name": "Meyk", "author_url": "", "post_date": "2018-05-07T19:02:02.360000", "content": "I think that feature engineering is needed ;) ... or just a simple feature to be honest. As one of kagglers mentioned (I am deeply sorry I've just remember your keynote not a nickname :( ) withdrawal of uploading kernels ~1 week before deadline will solve the case.
", "votes": 2, "replies": [] }, { "id": 324514, "author_name": "Samrat Pandiri", "author_url": "", "post_date": "2018-05-07T19:02:09.223000", "content": "@authman Disqualifying only that entry makes sense.. But already there are blends of blends with a much higher score!!
", "votes": 3, "replies": [] }, { "id": 324515, "author_name": "yimacs", "author_url": "", "post_date": "2018-05-07T19:02:58.637000", "content": "+1
", "votes": 0, "replies": [] }, { "id": 324517, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-07T19:05:42.230000", "content": "@ authman
\n\nI will leave it to Kaggle to decide what they want to do about this.
\n\nI don't have a medal yet. I got from 0.9797 to 0.9800 in my very last submission today even as all this was unfolding. Yet I chose NOT TO USE the 0.9811 file.
\n\nI am only questioning the idea of using a fully developed output file to improve one's score.
\n\n:)
\n\nCheers\nShanth
", "votes": 4, "replies": [] }, { "id": 324518, "author_name": "inversion", "author_url": "", "post_date": "2018-05-07T19:07:04.450000", "content": "Hi @Shanth -
\n\nI see where there was confusion in my response. I was responding directly to your question about having a private kernel. If your kernel is private, you cannot share it, or the output of it, with anyone that is not on your team.
\n\nPublicly-shared kernels, and the output to such, are available to anyone. This creates the current difficulty.
", "votes": 1, "replies": [] }, { "id": 324521, "author_name": "Shitian Ni", "author_url": "", "post_date": "2018-05-07T19:11:03.833000", "content": "@inversion In Data Science Bowl 2018, admins could delete people from leaderboard if āthey are against the spirit of the competitionā. Can kaggle do the same here? I think now itās the time.
", "votes": 6, "replies": [] }, { "id": 324537, "author_name": "Rui Li", "author_url": "", "post_date": "2018-05-07T19:24:17.687000", "content": "I would agree with @authman.
\n\nI think disable submissions (except who originally posted it on kernels) that are completely identical to the publicly available csv seems to be a better idea. Otherwise, if posting csv is considered illegal sharing, then a lot of hard working kernel contributors will be disqualified, since many of the kernels have posted the results (just not as high as 0.9811 or as late in the competition).
\n\nAlthough the situation might be very frustrating, we should be very cautious about disqualifying kagglers from competition, especially if the rule is not well stated, emphasized and enforced before. I think any rule change or re-clarification and enforcement should be done for future competition with a clear statement at the beginning of the competition. (But this is just my opinion and up for discussion).
\n\nOtherwise, Kaggle can wipe out half of the lead board with a snap of his fingerā¦
", "votes": 4, "replies": [] }, { "id": 324544, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-07T19:28:52.580000", "content": "@ inversion
\n\nThanks for the response. Hope you can speak with the rest of the Kaggle team to consider possible options.
\n\nYour response is completely understandable if someone just ran the kernel and generated their own outputs - I guess there would be no way to stop or monitor it.
\n\nBut having a public kernels output AS A DATA SOURCE and then submitting it just doesn't seem right. It is clearly against the spirit of the competition and the platform that Kaggle is.
\n\nHope the Kaggle team can do something about it .
\n\nRegards\nPrasanth
", "votes": 1, "replies": [] }, { "id": 324553, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-07T19:34:40.670000", "content": "@ RLstat - I Like the Thanos reference :)
", "votes": 1, "replies": [] }, { "id": 324607, "author_name": "Daniel J Brooks", "author_url": "", "post_date": "2018-05-07T21:09:01.270000", "content": "@Inversion\nEarly on kernels help newer Kagglers get up to speed. Right before submission, they provide an unfair advantage.
\n\nI strongly recommend disabling or delaying kernel submissions during the last week. Competitors should be given a chance to think about the problem on their own.
", "votes": 1, "replies": [] }, { "id": 324608, "author_name": "FrLi", "author_url": "", "post_date": "2018-05-07T21:09:05.783000", "content": "I am thinking do we really need kernel for these competitions. For people who are really working on the problem, discussion is enough with no need showing all code. For new comers like me, kernels in those competitions with no money involved should already be enough.
", "votes": 2, "replies": [] }, { "id": 325064, "author_name": "Mayank Soni", "author_url": "", "post_date": "2018-05-08T04:43:08.823000", "content": "Inversion : I gave it a thought and here are my 2 cents. \"Like there is a team merger deadline , let's have a kernel submission deadline as well (Maybe same as team merger) \" That way everyone gets to learn as well and if a competition lasts for 2 months , i think 6 weeks are enough to generate ideas in kernels. That way this level of frustration can be avoided.
", "votes": 0, "replies": [] }, { "id": 325101, "author_name": "happycube", "author_url": "", "post_date": "2018-05-08T05:55:04.120000", "content": "A couple of rules that'd make sense to me:
\n\n\n\n\nPlease continue to post your thoughts and suggests in this thread. Your frank feedback is very much appreciated.
\n
One small suggestion:
\n\nWith data this size, tied scores are nearly certain to be identical CSVs.
\n\nBreaking ties by time of submission becomes less meaningful/appropriate the more people submit a publicly shared CSV.
\n\nWhy not group identical entries into one implicit team rank?
\n\nIf you keep the original number of teams for medals calculations (as in two stage competitions) that would move some teams upwards into the medals zones. (Actually it may reward some admirable people - those who worked hard and gained an honest score just a hair below the shared solution but still did not use it.)
\n\nPerhaps even dilute the points of the new merged teams in the same way actual teams are penalized by size. (More frankly though: I don't see why simply submitting someone else's CSV file - shared publicly or privately - should lead to any kind of reward/recognition at all. It reminds me of this article by Charlie Brooker.)
\n\nThis would not address the people who blended or made tweaks to the CSV... but in the past nothing at all is done to fix last minute sharing. Merging implicit teams would be a small step up from that.
", "votes": 6, "replies": [] }, { "id": 325658, "author_name": "YoungLamb", "author_url": "", "post_date": "2018-05-08T17:08:58.043000", "content": "That is a reasonable move, people who having exactly same submission count as one team and they would get few points. I think kaggle team needs to do something to those identical submissions, otherwise it could easily happen again when someone else wants to have some fun (like Dirk) and creates a new account , sharing high-rank solution at the final day..... This would be extremely discouraging for majority of competitors, if no penalty applied on it.
", "votes": 3, "replies": [] }, { "id": 326480, "author_name": "YUNFEI DUAN", "author_url": "", "post_date": "2018-05-09T19:48:45.367000", "content": "If kaggle stays like this, let the copies win, then most of the medals and ranks are valueless.\nThink about that some new platforms focused on truely fair competitions come out. Kaggle is ruining itself.
", "votes": 2, "replies": [] }, { "id": 326517, "author_name": "Jihye Sofia Seo", "author_url": "", "post_date": "2018-05-09T21:20:44.323000", "content": "Are there already some new platforms?
", "votes": 0, "replies": [] }, { "id": 326518, "author_name": "Jihye Sofia Seo", "author_url": "", "post_date": "2018-05-09T21:21:50.680000", "content": "Agreed. Kernels should become public only AFTER competition maybe.
", "votes": -1, "replies": [] }, { "id": 326521, "author_name": "Callum Gundlach", "author_url": "", "post_date": "2018-05-09T21:28:52.513000", "content": "Kernels are very helpful for beginners and sharing contributions. I don't agree with the idea that they should not be public, however they should be limited the final week of competition.
", "votes": 0, "replies": [] }, { "id": 326543, "author_name": "YUNFEI DUAN", "author_url": "", "post_date": "2018-05-09T22:32:29.967000", "content": "@Jihye topcoder has a data science channel.
", "votes": 0, "replies": [] }, { "id": 326545, "author_name": "YUNFEI DUAN", "author_url": "", "post_date": "2018-05-09T22:38:02.520000", "content": "Just block the result csv. I learn from kernels, not csv. What is the csv using for? To prove the kernel is true?
", "votes": 0, "replies": [] }, { "id": 326549, "author_name": "YUNFEI DUAN", "author_url": "", "post_date": "2018-05-09T22:47:31.197000", "content": "@Jihye and NUMERAI
", "votes": 0, "replies": [] } ] }, { "id": 324772, "author_name": "rocuku", "author_url": "", "post_date": "2018-05-07T23:44:51.607000", "content": "I worked hard for weeks to get score 0.9810, hoping to win my first silver medal, or at least a bronze medal
\n\nnow I just want to cry...
", "votes": 29, "replies": [] }, { "id": 324392, "author_name": "Samrat Pandiri", "author_url": "", "post_date": "2018-05-07T16:56:54.173000", "content": "Spent many weeks and weekends literally and was in the top 5% until late in the evening. Now I see that I'm in the border of loosing my first medal that I was dreaming from the last 2 months.. \nNot sure what to say :( \nWish some one can understand the pain!!!
\n\nSamrat : i understand your position . i have felt the same way . let's hope for the best in PB.
", "votes": 1, "replies": [] }, { "id": 324418, "author_name": "Samrat Pandiri", "author_url": "", "post_date": "2018-05-07T17:22:29.510000", "content": "@mayanksoni The huge difference will definitely push me down... In 1 hour I dropped almost 130 positions..
", "votes": 1, "replies": [] }, { "id": 324424, "author_name": "Joe Eddy", "author_url": "", "post_date": "2018-05-07T17:26:20.973000", "content": "Whatever you choose now, you should be proud of what you accomplished with honest effort. In my opinion, what you learn from putting in that effort is much more valuable than your competition placement - both in terms of your broader ML/data science skillset and your ability to succeed in future kaggle competitions.
\n\nI hope that you won't feel discouraged from competing in future competitions - I'm sure that your results will only get better and better. In my first competition (Instacart), I crawled into the top 10% before plummeting when 50th place decided to open source their solution in the last week. But I learned a lot there and it helped me in more recent competitions. Kaggle success is a marathon, not a sprint.
", "votes": 8, "replies": [] }, { "id": 324713, "author_name": "Eric", "author_url": "", "post_date": "2018-05-07T22:45:48.277000", "content": "Samrat, when I look at top kagglers, I understand they are not here for one competition but they have done many... and will continue to do many... At the end of the day, the score is less important than your real ability to solve thanks to machine learning a problem. This is what is going to be useful in your day to day job... Not the medals that are just good for flaunting one's ego
", "votes": 3, "replies": [] }, { "id": 324762, "author_name": "", "author_url": "", "post_date": "2018-05-07T23:33:16.907000", "content": "", "votes": 0, "replies": [] }, { "id": 324854, "author_name": "marketneutral", "author_url": "", "post_date": "2018-05-08T00:43:13.933000", "content": "Public leaderboard justice for you. Congratulations.
", "votes": 3, "replies": [] } ] }, { "id": 324658, "author_name": "Daniel J Brooks", "author_url": "", "post_date": "2018-05-07T22:00:59.230000", "content": "This is my first competition.
\n\nMy goal is to earn a bronze medal. The new kernel pushed my submission, and others, out of medal range.
\n\nThe best medal winning strategy, currently, is to copy and submit the highest scoring public kernel.
\n\nI've decided to submit my own solution, rather than the public model.
\n\nI hope that there is a policy change to address game breaking kernel submissions in the last days of the competition.
\n\nI quite like the Kaggle platform and community, but it's hard to justify spending weeks on a problem to compete with copy and paste.
", "votes": 22, "replies": [ { "id": 324680, "author_name": "Matthew Anderson", "author_url": "", "post_date": "2018-05-07T22:15:30.360000", "content": "Hey, maybe you'll beat his model on the private LB. This doesn't happen for most competitions.
", "votes": 2, "replies": [] }, { "id": 324850, "author_name": "Callum Gundlach", "author_url": "", "post_date": "2018-05-08T00:41:58.313000", "content": "I'm actually in the same position and feel the same way
", "votes": 3, "replies": [] } ] }, { "id": 324220, "author_name": "CPMP", "author_url": "", "post_date": "2018-05-07T12:26:48.227000", "content": "I fully agree, and I wrote a similar post in Quora competition, see https://www.kaggle.com/c/quora-question-pairs/discussion/33801
\n\nI bet others have also written about the same issue in other competitions.
\n\nTo be clear, I am not against sharing high value kernels. I am against sharing high value kernels near the end of a competition. I personally would like that no new kernel sharing is possible in the last week.
", "votes": 22, "replies": [ { "id": 324227, "author_name": "", "author_url": "", "post_date": "2018-05-07T12:49:57.737000", "content": "Yes, this topic comes up at nearly every competition deadline. My question is: what prevents you from blending with \"high value kernels\"? Just save a couple of submits before the final deadline and you are good.
", "votes": -5, "replies": [] }, { "id": 324234, "author_name": "CPMP", "author_url": "", "post_date": "2018-05-07T13:08:03.237000", "content": "\n\n\nwhat prevents you from blending with \"high value kernels\"?
\n
My job ;)
\n\nOther reasons can be exams, travel, family events, whatever that can prevent you from working on a Kaggle competition for few days in a row.
\n\nWhatever the reason, it may happen that you cannot work on a competition the last day. If someone shares something valuable when one cannot react to it, then it is unfair IMHO.
\n\nFor me, the only way to counter it is to take a day off, like I'm doing today. But this is not always doable.
\n\nBut your question is valuable. That's why I am not against sharing in general, as one can always try to benefit from what has been shared. I guess this is your point, and I agree with it.
", "votes": 12, "replies": [] }, { "id": 324249, "author_name": "", "author_url": "", "post_date": "2018-05-07T13:29:25.733000", "content": "People who do Kaggle full time (as myself) always have an advantage over those who also have a job. Some people, I assume, work in teams from one Kaggle profile which gives even more advantage. The problem of allocating your free time is related to opportunity costs. And it is totally irrelevant to the question of sharing the high-value kernels at the end of the competition.
", "votes": 1, "replies": [] }, { "id": 324251, "author_name": "Kevin", "author_url": "", "post_date": "2018-05-07T13:31:04.847000", "content": "\n\n\nPavel Pleskov wrote
\n \n\n\nYes, this topic comes up at nearly every competition deadline. My question is: what prevents you from blending with \"high value kernels\"? Just save a couple of submits before the final deadline and you are good.
\n
Computational resources ;)
", "votes": 0, "replies": [] }, { "id": 324254, "author_name": "Matthew Hendricks", "author_url": "", "post_date": "2018-05-07T13:36:31.170000", "content": "Blending someone else's work seems to be against the spirit of the competition and the ranking system. You can achieve a very high rank without knowing much about coding or machine learning. Doesn't seem right to me.
", "votes": 6, "replies": [] }, { "id": 324258, "author_name": "CPMP", "author_url": "", "post_date": "2018-05-07T13:46:55.193000", "content": "@Pavel
\n\n\n\n\nSome people, I assume, work in teams from one Kaggle profile which gives even more advantage.
\n
I agree with you on that one. For the rest, let's agree that we disagree ;)
", "votes": 3, "replies": [] }, { "id": 324265, "author_name": "", "author_url": "", "post_date": "2018-05-07T13:57:28.220000", "content": "@Matthew I would love to see someone from top-100 (which is a very high rank for me) without knowing much about coding or machine learning :) IMHO getting the best possible score is indeed the spirit of the competition. What is wrong with finding creative ways to do it based on other people work? Isn't it how teamwork goes? By the way, have you ever tried to blend kernels and get the high score on a private leaderboard? Not that easy to do it consistently, trust me
", "votes": 5, "replies": [] }, { "id": 324269, "author_name": "Mayank Soni", "author_url": "", "post_date": "2018-05-07T14:03:32.927000", "content": "Different TimeZone , Work that pays , Family ...
", "votes": 2, "replies": [] }, { "id": 324271, "author_name": "", "author_url": "", "post_date": "2018-05-07T14:04:13.343000", "content": "@Kevin there is always this complaint about lack of resources. Not enough time, experience, GPUs, cores, RAM... Yet some people still win without having all of that and not making excuses.
", "votes": 0, "replies": [] }, { "id": 324272, "author_name": "Matthew Hendricks", "author_url": "", "post_date": "2018-05-07T14:05:32.807000", "content": "right now you can run a public kernel and score in the top 150 without knowing anything about machine learning. @Pavel you should be upset about that because it makes your high ranking virtually meaningless
", "votes": 1, "replies": [] }, { "id": 324281, "author_name": "", "author_url": "", "post_date": "2018-05-07T14:17:23.970000", "content": "Right tail of the distribution never gets upset by the left tail :)\nPS: what if I tell you that there is also a private LB...
", "votes": 0, "replies": [] }, { "id": 324283, "author_name": "CPMP", "author_url": "", "post_date": "2018-05-07T14:23:00.967000", "content": "@Mathew, let's see the private LB before we can discuss ranks. Also, what is the public kernel that makes you in top 150 on the public LB? It needs to score at least 0.9811, and I don't remember seeing one.
", "votes": 0, "replies": [] }, { "id": 324289, "author_name": "Matthew Hendricks", "author_url": "", "post_date": "2018-05-07T14:29:13.047000", "content": "So you're not upset if a left-tailer can masquerade as a right-tailer by using quality public kernels posted on the last day? If you score in the top 150 in every competition by simply running a good kernel on the last day, wouldn't you be ranked among the top 100 overall? I think the ranking system is flawed but I am not sure how to fix it.
", "votes": 1, "replies": [] }, { "id": 324290, "author_name": "Arthur Llau", "author_url": "", "post_date": "2018-05-07T14:31:47.187000", "content": "@CPMP He is referring to this one.
", "votes": 2, "replies": [] }, { "id": 324291, "author_name": "Matthew Hendricks", "author_url": "", "post_date": "2018-05-07T14:32:11.217000", "content": "Yes there is one that scores 0.9811. People have been complaining about it this morning.
", "votes": 2, "replies": [] }, { "id": 324302, "author_name": "CPMP", "author_url": "", "post_date": "2018-05-07T14:48:57.157000", "content": "My bad, I didn't see it. Will look at it right now!
\n\nThis sharing so late is silly.
\n\nEdited: no low hanging fruit that I can reuse in it, will stick to my plan for the next few hours. That's why sharing so late is silly, people don't have time to react to it. I feel for all those who put hard work and will be displaced by people who happen to have machines large enough to run this before competition ends. The only counterpart is that this kernel may overfit to the public LB (like my code now that I think of it ... )
", "votes": 1, "replies": [] }, { "id": 324306, "author_name": "Meyk", "author_url": "", "post_date": "2018-05-07T14:52:51.717000", "content": "\n\n\nCPMP wrote
\n \nI agree with you on that one. For the rest, let's agree that we disagree ;)
\n
'Men in black' is quite enjoying movie
", "votes": 1, "replies": [] }, { "id": 324307, "author_name": "Arthur Llau", "author_url": "", "post_date": "2018-05-07T14:53:42.430000", "content": "And you even don't have to run it ...
", "votes": 3, "replies": [] }, { "id": 324361, "author_name": "", "author_url": "", "post_date": "2018-05-07T16:18:50.977000", "content": "@Matthew according to my ranking score tool https://www.kaggle.com/ppleskov/ranking-comparison-tool 150th place gives you around 1500 points, so to reach the top-100 score of 38000 you will need to perform this trick in 25 competitions. It is highly unlikely to do so in a reasonable time since not in every competition somebody shares high results on the last day and there are around 20 competitions per year. I'm not even talking about reaching the very top of 180000 points. So yes, it does not bother me at all.
", "votes": 4, "replies": [] }, { "id": 324375, "author_name": "Matthew Hendricks", "author_url": "", "post_date": "2018-05-07T16:28:53.260000", "content": "Wow. Someone with no knowledge of machine learning could theoretically be ranked in the top 100 on Kaggle within just a few years. That bothers me and I haven't invested time in acquiring a high rank. I think it's just unfair to those that have invested a lot of time to signal their skills with a high rank.
", "votes": 2, "replies": [] }, { "id": 324383, "author_name": "CPMP", "author_url": "", "post_date": "2018-05-07T16:42:58.563000", "content": "@Mathew, The issue does not impact top performers, see Pavel's reaction above. I agree with him that there is no way someone can make it in top 150 in the global ranking system just by using public kernel outputs. No way.
\n\nThat's not why this sharing is an issue. It is an issue because of all whose ranks will be negatively impacted in this competition may not want to invest time in the next competition.
", "votes": 5, "replies": [] }, { "id": 324398, "author_name": "Matthew Hendricks", "author_url": "", "post_date": "2018-05-07T17:03:57.547000", "content": "@CPMP I am new so don't know how often this happens. So this late posting doesn't happen often? Others seem to think it happens all the time. In any event, it would be easy to stop if Kaggle just blocked new kernel posts a few days before the end of a competition.
", "votes": 0, "replies": [] }, { "id": 324444, "author_name": "James Trotman", "author_url": "", "post_date": "2018-05-07T17:52:30.817000", "content": "Check out the excellent Scripty McScriptface the Lazy Kaggler analysis by dune_dweller.
\n\nFor the first 15 months or so of Scripts/Kernels, submitting the best public script could get you to\n13881 points, at that time good for 449th globally...
\n\nIt would be fantastic if Kaggle updated the Meta Kaggle data set & that analysis kernel could be re-run on all the competitions to date...
", "votes": 3, "replies": [] }, { "id": 324454, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-07T17:56:32.990000", "content": "Yeah DuneWeller is awesome - the good news from this was that at least they removed the automated submission button - zoiks! If that was still there my dog would come in the top 500
", "votes": 2, "replies": [] }, { "id": 324677, "author_name": "Eric", "author_url": "", "post_date": "2018-05-07T22:13:06.147000", "content": "Agree that this kind of late kernel does not harm top performers. This is bad that it is shared at the last minute but it is usefull for people like me who are at the learning stage to see imagination of other people.
", "votes": 2, "replies": [] }, { "id": 325138, "author_name": "Stephen McInerney", "author_url": "", "post_date": "2018-05-08T06:54:05.007000", "content": "Useful to preserve the plaintext names of both the submitter and kernel, since the kernel was removed and the submitter seems to have been disqualified:
\n\nKernel: www.kaggle.com/midi303/entire-dataset-lb-0-9811-lightgbm
\n\nSubmitter: Dirk www.kaggle.com/midi303
\n\n(I'm pretty sure this is still public knowledge, since it's in each of our inboxes if you subscribed to email delivery)
", "votes": 0, "replies": [] } ] }, { "id": 324831, "author_name": "Sun ZhiHao", "author_url": "", "post_date": "2018-05-08T00:31:09.353000", "content": "It is so bad to see silver region from 153-171 all share one kernel \"Merge and avg.\" and gain silver medal with less than 10 submissions. I suggest committee to cancel grade for all participants with exactly same submission. This file is super large and exact same submission means plagirism
", "votes": 19, "replies": [ { "id": 324945, "author_name": "Matthew Anderson", "author_url": "", "post_date": "2018-05-08T02:03:18.237000", "content": "This is the actual definition of harmful plagiarism, which shouldnāt be allowed on Kaggle as it isnāt allowed in any other academic setting.
", "votes": 4, "replies": [] }, { "id": 325003, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-08T03:08:11.993000", "content": "Well said Sun Zhiaho
", "votes": 0, "replies": [] }, { "id": 325268, "author_name": "Antonis Maronikolakis", "author_url": "", "post_date": "2018-05-08T09:12:43.170000", "content": "Plagiarism is a heavy term and I don't like seeing it thrown around lightly. In Kaggle it is within the rules to use public kernel submissions, so this is not plagiarism since plagiarism revolves around \"stealing\" someone else's work and claiming it as your own. Under Kaggle rules, submitting a public kernel submission is within the rules so this is by definition not plagiarism.
\n\nI would like to see a solution that properly resolves this as much as anyone else on here, but throwing terms and accusations around is not conductive to the discussion.
", "votes": 1, "replies": [] }, { "id": 325606, "author_name": "Matthew Anderson", "author_url": "", "post_date": "2018-05-08T16:12:43.070000", "content": "Anthony, I agree that blind accusations are wrong, but exactly resubmitting the output of other peoples' work without even running the kernel itself does seem wrong and appears to be very similar to plagiarism. Plagiarism, as in the practice of taking someone else's work or ideas (kernels) and passing them off as one's own (on the LB).
", "votes": 0, "replies": [] } ] }, { "id": 325065, "author_name": "Pavel (Pasha) Gyrya", "author_url": "", "post_date": "2018-05-08T04:43:33.110000", "content": "It seems as results stand currently, it's at least 30% of medals are for exact copied work without any additions at all. \nPerhaps even more people leveraged late solutions in their blends. This system promotes people to think less, not more. I wonder how many of these folks even understand what they copy - we don't have a way to tell.
\n\nI wonder if people who submit these solutions really want to showcase such a result on resume... This is such a shame show. To discourage it, I could suggest Kaggle to
\n\nI would fully agree for restricted sharing in the end of the competition as well. Typically people focus on working on their solutions at this time, not everyone has a chance to react to public kernel in a short period of time even if they wanted to.
\n\nI think also it makes sense for Kaggle to retro-actively adjust results of this competition itself to help recognize the modeling work that people put in.
\n\nHope this helps figure out a good approach that promotes excellence and collaboration, while giving credit where it's due.
", "votes": 16, "replies": [ { "id": 325164, "author_name": "Izmaylov Konstantin", "author_url": "", "post_date": "2018-05-08T07:23:23.557000", "content": "Great idea! But people can blend anyway, there will be kernels of how to blend with different random seeds.
", "votes": 2, "replies": [] } ] }, { "id": 324416, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-07T17:21:25.490000", "content": "Why would recruiters use kaggle now that these blinking kernels exist. Kaggle sort it out - you will lose hard working competitors too! When I wrote the kernels schmernels post I was inaccurate when I said start two weeks from the end - apologies ;(
", "votes": 15, "replies": [] }, { "id": 325412, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-08T11:48:42.450000", "content": "One thing to not forget - if that kernel hadn't been released - this was a very very successful competition overall - lots of scope for feature engineering that didn't have leakage!
", "votes": 11, "replies": [] }, { "id": 324393, "author_name": "Pranav Pandya", "author_url": "", "post_date": "2018-05-07T16:58:12.113000", "content": "It's very disappointing to see LB being destroyed in last hours. I am sure that Kaggle can't do anything about it because they will never know who will be posting what and when. But at least they can have some mechanism to allow community to punish such D***s flying around in every competition in last hours.
This is regular problem and needs a solution. I and many others strongly recommend having downvote option for kernels .
\n\nMany kagglers have already educated that novice kernel author but it seems that he is purely in the mood of destroying leaderboard and not taking down that kernel. So may be there is a need to have a Hall of Shame post or a kernel like this
", "votes": 12, "replies": [ { "id": 324681, "author_name": "Eric", "author_url": "", "post_date": "2018-05-07T22:15:48.307000", "content": "All, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel. I understand this can be seen like cheating for people who are using this kernel to increase their score but for top performers, they are way above this 0.9811 so this should not be such an issue on the kaggle forum...
", "votes": -4, "replies": [] }, { "id": 324686, "author_name": "Joe Eddy", "author_url": "", "post_date": "2018-05-07T22:18:42.273000", "content": "\n\n\neric wrote
\n \n\n\nAll, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel. I understand this can be seen like cheating for people who are using this kernel to increase their score but for top performers, they are way above this 0.9811 so this should not be such an issue on the kaggle forum...
\n
It's not ironic at all. Many of us care about the broader community beyond ourselves and want to see diligent, honest work rewarded over last minute button pressing. Also, many of us have been in this exact position in the past and understand how frustrating it is, so can empathize.
", "votes": 12, "replies": [] }, { "id": 324701, "author_name": "Eric", "author_url": "", "post_date": "2018-05-07T22:34:49.127000", "content": "Joe Eddy, thanks for being so compassionate about poor kagglers like me. But I still believe the real issue is to remember that Kaggle has been created to make learning data science and machine learning a game and that in any game the most important stuff at least for me is to participate and have some fun. Then, if you win, this is even better but winning should not be seen as the ultimate goal
\n\nAnd I cannot resist to quote a famous French countryman\n\"The most important thing in the Olympic Games is not winning but taking part; the essential thing in life is not conquering but fighting well.\" - Pierre de Coubertin
", "votes": -7, "replies": [] }, { "id": 324728, "author_name": "Pranav Pandya", "author_url": "", "post_date": "2018-05-07T23:01:14.760000", "content": "\n\n\neric wrote
\n \n\n\nAll, it is quite ironic that people that are ranked well above the LB 0.9811 complain about this kernel.
\n
When you're part of the community, you can feel the pain and would stand up naturally.
\n\nI have faced such last day disasters in some of my previous competitions and can totally relate how it feels when few people (ya, just few out of thousand participants) destroys the leaderboard.
", "votes": 3, "replies": [] }, { "id": 324732, "author_name": "Eric", "author_url": "", "post_date": "2018-05-07T23:04:22.313000", "content": "Thanks Pranav for this nice feedback! Understood then
", "votes": 1, "replies": [] } ] }, { "id": 324358, "author_name": "Cong", "author_url": "", "post_date": "2018-05-07T16:13:09.130000", "content": "within last hour, many with <5 subs jumps to this score. Now imagine you were someone having 9810 last night and thought you could get your first silver medal.
", "votes": 9, "replies": [] }, { "id": 325047, "author_name": "Marvin", "author_url": "", "post_date": "2018-05-08T03:59:36.907000", "content": "as a new player of kaggle, I just want to cry. Very disappointed.
", "votes": 8, "replies": [ { "id": 326548, "author_name": "YUNFEI DUAN", "author_url": "", "post_date": "2018-05-09T22:44:37.227000", "content": "I feel the same. I feel maybe I came to a wrong place, or I should learn something shame rather than real knowledge if I want a medal.
", "votes": 0, "replies": [] } ] }, { "id": 325027, "author_name": "trphng", "author_url": "", "post_date": "2018-05-08T03:32:22.803000", "content": "Very bad Kaggle experience this morning.
", "votes": 7, "replies": [] }, { "id": 324411, "author_name": "Matthew Hendricks", "author_url": "", "post_date": "2018-05-07T17:17:50.487000", "content": "It is interesting to see how many people are willing to use the infamous late kernel to obtain a now meaningless high rank in this competition. Maybe that should be the next competition: predict which Kagglers will make the late submission of someone else's work.
", "votes": 8, "replies": [ { "id": 324539, "author_name": "Meyk", "author_url": "", "post_date": "2018-05-07T19:24:29.323000", "content": "I like the idea. We will have training set as well :D
", "votes": 3, "replies": [] }, { "id": 324994, "author_name": "Wenjie Bai", "author_url": "", "post_date": "2018-05-08T02:52:09.337000", "content": "I am looking forward to join you in that competition.
", "votes": 0, "replies": [] } ] }, { "id": 324384, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-07T16:44:40.627000", "content": "It is worse now there are datasets - I have been warning about this for months - you just get a \"we can't do anything about it\" which is something I would fire someone over if they worked for me (I am a bad boss I suppose)
", "votes": 8, "replies": [] }, { "id": 324425, "author_name": "Izmaylov Konstantin", "author_url": "", "post_date": "2018-05-07T17:26:29.517000", "content": "I think, that this topic has made 0.9811 kernel much more popular.
", "votes": 5, "replies": [] }, { "id": 324405, "author_name": "Icfstat", "author_url": "", "post_date": "2018-05-07T17:10:04.963000", "content": "I went to sleep in silver medal position, then I woke up and found that infamous kernel. I have made a significant effort and I was enjoying the competition, now this situation is very frustrating. Now ,as many competitors, I am at work and don't have the time and resources to train more models. So, kaggle user who want to use this kernel consider the fairness of the competition and don't use it please.
", "votes": 6, "replies": [] }, { "id": 324345, "author_name": "A.Barqawi", "author_url": "", "post_date": "2018-05-07T15:54:26.423000", "content": "Agree with you, investing big time and the last day you find high score Kernal where no time to learn from it or maybe you don't have submissions left to use it ! then hours of your afford vanish.
\n\nMy suggestion after the rules acceptance deadline to stop public Kernals because no much time to learn from it. OR at least stop upload input to public Kernal at that date
\n\nAlso regarding Kaggle removing scores of competitors because of cheating usually happen for unaware of rules to publish results primitively or share it with friend, I suggest Kaggle make a run to remove them at acceptance deadline, As people decide to invest more and more time on competition after it or could decide to rent a server and pay money then find score vanished by the end. Also more fair to have time to send appeal in case required.
", "votes": 4, "replies": [] }, { "id": 324767, "author_name": "eagle4", "author_url": "", "post_date": "2018-05-07T23:38:30.560000", "content": "I want to resign from this competition.
\n\nYes, the ranks matter (it is a competition!) and yes I m pissed that my participation is going to benefit last minute public blending competitors!
", "votes": 5, "replies": [] }, { "id": 324991, "author_name": "SubikashPal", "author_url": "", "post_date": "2018-05-08T02:47:41.170000", "content": "Is it cheating or unethical? I am completely against the term cheating here. There's no rule in the competition that describes public sharing at the last moment is not allowed and will be termed as \"Cheater\"
\n\nCouple of cases -
\n\n\n\n\nSomebody uses the csv to upload. -- Is it against rule?\n Somebody blended with their model-- is it cheating?\n Somebody uses some of their features and used it ( like me) in their model. -- what would you call it?\n Somebody who did their blending at last moment with/without using this kernel's CSV. -- can it be tracked?
\n
Somehow I would term the entire episode as Unethical rather than cheating. I am not sure what Kaggle is going to do today but its unfair to remove all those who directly/indirectly uses the kernel. Rather as already mentioned , Kaggle should do something in future competition to stop the practice.
", "votes": 3, "replies": [ { "id": 325000, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-08T03:05:50.283000", "content": "@ Subiksha Pal
\n\nYou used someone else's output. If you did that in a University you would get expelled/disbarred. \nIf you did that in a company you would be hit with a class action lawsuit for intellectual property theft. Perhaps you should bear that in mind before defending your position. What is really disturbing is how people are actually defending flagrant PLIAGARISM
\n\nI have no problems with blending or taking ideas from someone else or even last minute kernel releases but even that needs to be fair. You can't Float the ANSWERS around for everybody to use.
", "votes": 0, "replies": [] }, { "id": 325013, "author_name": "SubikashPal", "author_url": "", "post_date": "2018-05-08T03:20:24.823000", "content": "I am not defending here . We all are here talking about this fairness. But how to define that? My perception was to have some rule to confine the publication of the kernel at last minutes.These all are unethical practice that I am trying to mention and should be stopped.
", "votes": 0, "replies": [] }, { "id": 325262, "author_name": "Antonis Maronikolakis", "author_url": "", "post_date": "2018-05-08T09:06:17.700000", "content": "@Shanth: I am as upset by this behavior as anyone, but this is in no way cheating since it broke no rules. It is not IP theft, since the guy shared the results publicly (and I'm sure kernels are published under some license, but I can't be bothered to check). In Universities you are explicitly not allowed to share exams/assignments, but in Kaggle this is encouraged by public kernels, so this is not a correct analogy as well.
\n\nIt is an ethical issue where users released competition-breaking content in the last second. Everything was within the rules.
", "votes": 1, "replies": [] }, { "id": 325320, "author_name": "Shanth", "author_url": "", "post_date": "2018-05-08T09:58:54.017000", "content": "@ Anthony Marakis -
\n\nLet's put this into context here a little bit. The problem we were working on needed extra computing power and memory optimizations to run the training models across the entire data set. Not everyone could do this for lack of resources. So, someone runs the model on all the data and understandably gets better results ( The 0.9811 kernel had features that were no different from the public kernels and got better results simply because it was using more data ) and decides to share the output file.
\n\nOthers, who have themselves not had access to such resources decide to USE THE OUTPUT FILE without even running a kernel and thats' OKAY? Hundreds of people UPLOAD THE SAME OUTPUT file and that's okay and that's NOT PLAGIARISM?
\n\nI don't think plagiarism is a heavy term. Plagiarism is when one copies someone else's work and passes it off as their own. Wouldn't this qualify ?\n :)
", "votes": -1, "replies": [] }, { "id": 325332, "author_name": "Antonis Maronikolakis", "author_url": "", "post_date": "2018-05-08T10:11:20.990000", "content": "I wholeheartedly agree that there is an issue (that's why I wrote this post :)). Actually, this is a huge issue. But plagiarism did not happen here. Plagiarism does not involve simple copying, but stealing. It is a type theft, which did not happen here.
\n\nThe only reason I don't want plagiarism to be thrown so easily around is because I know how terrible it can be for people involved. I was a part of a large community of writers, and there were cases where people made money plagiarizing (stealing licensed work and passing it off as their own) the work of others until they were caught and sued to oblivion. Even then, the damages to the writers were probably in the thousands of dollars, since they couldn't sell their own work, because someone else was claiming it was theirs (and takes a ton of time, money and effort to prove something is yours). It's a very nasty business, probably the nastiest in the creative community, and I don't want the term to start losing its meaning.
\n\nI prefer to see our efforts into fixing the actual issue, since this cannot continue. I am seriously considering not participating until this is fixed, even though I was one of the lucky ones who ended up right where they were before the shenanigans started.
", "votes": 2, "replies": [] } ] }, { "id": 325334, "author_name": "HuyenNguyen", "author_url": "", "post_date": "2018-05-08T10:12:31.390000", "content": "I would say no sharing of public kernels that can achieve top 20% score. It was disappointing I was in the top 6% last night. Got up to find 300 new people on top of me with the same score!! ARGH!!
", "votes": 3, "replies": [] }, { "id": 324655, "author_name": "pocket", "author_url": "", "post_date": "2018-05-07T21:59:56.930000", "content": "How about we post a 0.9832 solution? (but with random noise in the private hours)
\nJust kidding...
There were talks about blends spoiling the community, blending attempts at last minute of this competition has made this even worse.
", "votes": 4, "replies": [] }, { "id": 324559, "author_name": "Scirpus", "author_url": "", "post_date": "2018-05-07T19:40:31.953000", "content": "For all those punters wishing to sleep tonight click on the option to unsubscribe ;)
", "votes": 4, "replies": [] }, { "id": 324522, "author_name": "Radu Stoicescu", "author_url": "", "post_date": "2018-05-07T19:11:26.350000", "content": "Can I withdraw from the competition?
\n\nI already have two dark spots on my portfolio:
\n\nSanta Gift Matching Challenge - Infinite Probabilistic Improver was published when I was on holiday
\n\nMercedes-Benz Greener Manufacturing - for most of us it seems the Manatees decided the place on LB
\n\nNow this.
", "votes": 4, "replies": [] }, { "id": 324456, "author_name": "", "author_url": "", "post_date": "2018-05-07T18:00:27.043000", "content": "", "votes": 4, "replies": [] }, { "id": 324233, "author_name": "", "author_url": "", "post_date": "2018-05-07T13:00:08.803000", "content": "", "votes": 4, "replies": [ { "id": 324299, "author_name": "", "author_url": "", "post_date": "2018-05-07T14:45:31.627000", "content": "", "votes": 1, "replies": [] }, { "id": 324721, "author_name": "", "author_url": "", "post_date": "2018-05-07T22:50:29.727000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 324894, "author_name": "", "author_url": "", "post_date": "2018-05-08T01:22:50.803000", "content": "", "votes": 2, "replies": [] }, { "id": 324401, "author_name": "", "author_url": "", "post_date": "2018-05-07T17:06:48.270000", "content": "", "votes": 2, "replies": [] }, { "id": 325136, "author_name": "", "author_url": "", "post_date": "2018-05-08T06:51:36.480000", "content": "", "votes": 3, "replies": [] }, { "id": 324245, "author_name": "", "author_url": "", "post_date": "2018-05-07T13:25:09.623000", "content": "", "votes": 1, "replies": [] }, { "id": 325414, "author_name": "", "author_url": "", "post_date": "2018-05-08T11:49:37.673000", "content": "", "votes": 1, "replies": [ { "id": 325416, "author_name": "", "author_url": "", "post_date": "2018-05-08T11:54:20.373000", "content": "", "votes": 1, "replies": [] }, { "id": 325425, "author_name": "", "author_url": "", "post_date": "2018-05-08T12:00:02.633000", "content": "", "votes": 0, "replies": [] }, { "id": 325509, "author_name": "", "author_url": "", "post_date": "2018-05-08T13:40:35.570000", "content": "", "votes": 5, "replies": [] } ] }, { "id": 324529, "author_name": "", "author_url": "", "post_date": "2018-05-07T19:16:28.877000", "content": "", "votes": 1, "replies": [ { "id": 324540, "author_name": "", "author_url": "", "post_date": "2018-05-07T19:25:04.403000", "content": "", "votes": 3, "replies": [] }, { "id": 324545, "author_name": "", "author_url": "", "post_date": "2018-05-07T19:29:38.260000", "content": "", "votes": 2, "replies": [] }, { "id": 324548, "author_name": "", "author_url": "", "post_date": "2018-05-07T19:31:55.427000", "content": "", "votes": 7, "replies": [] }, { "id": 325148, "author_name": "", "author_url": "", "post_date": "2018-05-08T07:09:24.650000", "content": "", "votes": 2, "replies": [] } ] }, { "id": 325426, "author_name": "", "author_url": "", "post_date": "2018-05-08T12:01:10.650000", "content": "", "votes": 2, "replies": [] }, { "id": 325578, "author_name": "", "author_url": "", "post_date": "2018-05-08T15:26:43.573000", "content": "", "votes": 0, "replies": [] }, { "id": 324458, "author_name": "", "author_url": "", "post_date": "2018-05-07T18:01:26.300000", "content": "", "votes": 1, "replies": [ { "id": 324463, "author_name": "", "author_url": "", "post_date": "2018-05-07T18:09:38.557000", "content": "", "votes": 10, "replies": [] }, { "id": 324469, "author_name": "", "author_url": "", "post_date": "2018-05-07T18:18:04.077000", "content": "", "votes": 3, "replies": [] }, { "id": 324485, "author_name": "", "author_url": "", "post_date": "2018-05-07T18:31:18.660000", "content": "", "votes": 1, "replies": [] }, { "id": 324624, "author_name": "", "author_url": "", "post_date": "2018-05-07T21:31:23.193000", "content": "", "votes": 2, "replies": [] }, { "id": 324684, "author_name": "", "author_url": "", "post_date": "2018-05-07T22:17:49.543000", "content": "", "votes": 0, "replies": [] }, { "id": 325140, "author_name": "", "author_url": "", "post_date": "2018-05-08T06:58:44.020000", "content": "", "votes": 0, "replies": [] }, { "id": 325174, "author_name": "", "author_url": "", "post_date": "2018-05-08T07:31:15.853000", "content": "", "votes": 1, "replies": [] }, { "id": 325261, "author_name": "", "author_url": "", "post_date": "2018-05-08T09:04:14.973000", "content": "", "votes": 1, "replies": [] }, { "id": 325271, "author_name": "", "author_url": "", "post_date": "2018-05-08T09:14:01.637000", "content": "", "votes": 2, "replies": [] } ] }, { "id": 324771, "author_name": "", "author_url": "", "post_date": "2018-05-07T23:42:01.020000", "content": "", "votes": -6, "replies": [ { "id": 324774, "author_name": "", "author_url": "", "post_date": "2018-05-07T23:46:37.863000", "content": "", "votes": 4, "replies": [] }, { "id": 324777, "author_name": "", "author_url": "", "post_date": "2018-05-07T23:51:00.403000", "content": "", "votes": -10, "replies": [] }, { "id": 324811, "author_name": "", "author_url": "", "post_date": "2018-05-08T00:17:07.967000", "content": "", "votes": 0, "replies": [] }, { "id": 324958, "author_name": "", "author_url": "", "post_date": "2018-05-08T02:15:41.787000", "content": "", "votes": 8, "replies": [] }, { "id": 324982, "author_name": "", "author_url": "", "post_date": "2018-05-08T02:38:14.243000", "content": "", "votes": 2, "replies": [] }, { "id": 325201, "author_name": "", "author_url": "", "post_date": "2018-05-08T08:10:18.617000", "content": "", "votes": -2, "replies": [] }, { "id": 325231, "author_name": "", "author_url": "", "post_date": "2018-05-08T08:42:53.237000", "content": "", "votes": -2, "replies": [] }, { "id": 325494, "author_name": "", "author_url": "", "post_date": "2018-05-08T13:21:42.913000", "content": "", "votes": 3, "replies": [] }, { "id": 325512, "author_name": "", "author_url": "", "post_date": "2018-05-08T13:42:45.013000", "content": "", "votes": 5, "replies": [] } ] }, { "id": 324760, "author_name": "", "author_url": "", "post_date": "2018-05-07T23:29:33.723000", "content": "", "votes": -4, "replies": [ { "id": 325391, "author_name": "", "author_url": "", "post_date": "2018-05-08T11:14:33.300000", "content": "", "votes": 0, "replies": [] }, { "id": 325401, "author_name": "", "author_url": "", "post_date": "2018-05-08T11:35:48.653000", "content": "", "votes": 5, "replies": [] } ] }, { "id": 327851, "author_name": "", "author_url": "", "post_date": "2018-05-12T16:53:56.623000", "content": "", "votes": 0, "replies": [] }, { "id": 325660, "author_name": "", "author_url": "", "post_date": "2018-05-08T17:14:58.163000", "content": "", "votes": 0, "replies": [] }, { "id": 324490, "author_name": "", "author_url": "", "post_date": "2018-05-07T18:41:37.507000", "content": "", "votes": 0, "replies": [] }, { "id": 343430, "author_name": "", "author_url": "", "post_date": "2018-06-15T09:58:40.437000", "content": "", "votes": -2, "replies": [] }, { "id": 326171, "author_name": "", "author_url": "", "post_date": "2018-05-09T11:48:09.450000", "content": "", "votes": 0, "replies": [] }, { "id": 325574, "author_name": "", "author_url": "", "post_date": "2018-05-08T15:11:36.023000", "content": "", "votes": -9, "replies": [ { "id": 325585, "author_name": "", "author_url": "", "post_date": "2018-05-08T15:36:29.180000", "content": "", "votes": 3, "replies": [] }, { "id": 325597, "author_name": "", "author_url": "", "post_date": "2018-05-08T15:56:44.010000", "content": "", "votes": -1, "replies": [] } ] }, { "id": 325451, "author_name": "", "author_url": "", "post_date": "2018-05-08T12:31:42.117000", "content": "", "votes": 0, "replies": [] }, { "id": 324267, "author_name": "", "author_url": "", "post_date": "2018-05-07T14:00:31.583000", "content": "", "votes": 0, "replies": [ { "id": 324285, "author_name": "", "author_url": "", "post_date": "2018-05-07T14:25:01.680000", "content": "", "votes": 1, "replies": [] }, { "id": 324311, "author_name": "", "author_url": "", "post_date": "2018-05-07T15:03:12.877000", "content": "", "votes": 7, "replies": [] }, { "id": 324687, "author_name": "", "author_url": "", "post_date": "2018-05-07T22:19:04.657000", "content": "", "votes": -1, "replies": [] } ] } ], "raw_markdown_by_id": { "324213": "*I know that I will sound whiny, but I think this is an important conversation. Also, excuse the kinda click-bait-y title.*\n\nI am new to Kaggle, having participated in just one competition before this, but I have noticed a lot of the old-timers comment how public kernels pushed them away from the site. I am one of those people who believe public kernels are great and crucial to the website's ecosystem. They have helped newcomers such as myself a lot in getting a head start and understanding the proper direction a problem can be solved in.\n\nSo, for educational purposes, kernels are *fantastic*. It seems though, as with the rest of the internet, when imaginary points are involved people tend to get a bit carried away and post things just for the sake of gaining more points (an example of this is the plethora of carbon-cut blending kernels).\n\nNormally, I wouldn't care much about such things. I understand that all these kernels that are a basic fork of each other drown out quality kernels, but I don't think that's a huge issue (people who want to read quality kernels will simply look past all the fork-of-fork-of-blending posts).\n\n**But**, people sometimes forget that this is a competition first and foremost and a lot of us put a ton of work into these, only to find ourselves drop a hundred places or more because someone decided to share a high-scoring model. I understand that, again, this is generally not a huge issue (in fact, it may be positive) since it drives us forward and pushes us to dissect the new methods and improve upon them. The problem is when people start posting great solutions right at the end of the competition. There is no time left to work and learn from these new insights, so the only thing this accomplishes is a) get the posters some shiny points, and b) help people who do nothing but fork and submit.\n\nI am starting to understand why old-timers post about kernels pushing them away. This is not in the spirit of competition and is disappointing to see time and time again. I believe something needs to be done about this. Maybe add downvotes to kernels? This way people can voice their disappointment in a manner that will affect point-gamers. Not sure if or how this will work, but something needs to change.\n\nAlso, a lot of people posting this type of kernels are new to the site and may not know of this \"Kaggle etiquette\". Maybe a guide should be written informing people of this?\n\n---\n\n**TL;DR**: Something needs to be done about people posting high-scoring models at the end of competitions.\n\nApologies for the long post.\n\nEDIT: Good luck gals and guys, brace for impact.", "324898": "I know what we learned is much more important than the score, but many companies are using Kaggle as one of evaluation metrics for hiring data scientists. Kaggle itself has the job board, and many companies are posting jobs there.\n\nI think Kaggle should do something to keep the reputation unless these companies are seeking \"Blending Scientists\".", "324378": "Kaggle should disable kernel sharing during the last week of the competition. I think this is the most optimal solution.", "325437": "Anyone help me debug this? It looks quiet runnable for kaggleš\n\n import numpy as np\n import warnings\n from pandas.tseries.offsets import DateOffset\n\n def win_medals():\n '''\n Only works to silver and bronze, have fun!\n '''\n if Competition == 'kaggle':\n assert Remaning_hours_to_deadline <= DateOffset(hours=12), \"Too early man,it's not time!\"\n \n grab_a_coffee()\n click_kernels_button()\n csvs = download_highest_score_csv(n=10)\n submits = do_blending(csvs, random_state=np.random.randint(1,100))\n submit_predictions(submits)\n \n if good_luck:\n return SILVER\n else:\n return BRONZE\n else:\n pass\n \n if __name__ == 'main':\n medals = win_medals()\n if entire_running_time >= DateOffset(hours=4)\n warnings.warn('You are wasting too much time! You should be quicker, try next time use only brute force without brain!')", "324664": "Many people have suggested to block kernels in the last week. We have been saying this for ages. Why kaggle hasn't acted on this?\n\n I am curious on the reasoning that kernels are still allowed in the last week ... I have honestly not seen much opposition for this suggestion (if any). \n\nOn the contrary I think kaggle seems to want this. Kernels and discussions are gamified and you can get points/ranks from these elements . If someone is not very high in competitions or he/she targets to get master/grandmaster through kernels, he/she can unleash a high scoring kernel near the end to get points/votes.", "324476": "This is a discussion point that gets re-ignited every few months, and becomes more challenging as Kaggle continues to add more functionality to Kernels and Datasets.\n\nWe are following this closely, and will be having focused discussions internally on how we can both (a) continue to provide value-added tools to the data science community as well as (b) minimize the amount of frustration that might arise during competitions because of these tools.\n\nPlease continue to post your thoughts and suggests in this thread. Your frank feedback is very much appreciated.", "324772": "I worked hard for weeks to get score 0.9810, hoping to win my first silver medal, or at least a bronze medal\n\nnow I just want to cry...", "324392": "Spent many weeks and weekends literally and was in the top 5% until late in the evening. Now I see that I'm in the border of loosing my first medal that I was dreaming from the last 2 months.. \nNot sure what to say :( \nWish some one can understand the pain!!!\n\n - So should I cheat now? or forget it and just let my heart cry!! - **Choose the 1st option**\n - Not sure if I should do justice to my Team Name **\"Hoping My First Medal\"**", "324658": "This is my first competition. \n\nMy goal is to earn a bronze medal. The new kernel pushed my submission, and others, out of medal range.\n\nThe best medal winning strategy, currently, is to copy and submit the highest scoring public kernel.\n\nI've decided to submit my own solution, rather than the public model. \n\nI hope that there is a policy change to address game breaking kernel submissions in the last days of the competition. \n\nI quite like the Kaggle platform and community, but it's hard to justify spending weeks on a problem to compete with copy and paste.", "324220": "I fully agree, and I wrote a similar post in Quora competition, see https://www.kaggle.com/c/quora-question-pairs/discussion/33801\n\nI bet others have also written about the same issue in other competitions.\n\nTo be clear, I am not against sharing high value kernels. I am against sharing high value kernels near the end of a competition. I personally would like that no new kernel sharing is possible in the last week.", "324831": "It is so bad to see silver region from 153-171 all share one kernel \"Merge and avg.\" and gain silver medal with less than 10 submissions. I suggest committee to cancel grade for all participants with exactly same submission. This file is super large and exact same submission means plagirism", "325065": "It seems as results stand currently, it's at least 30% of medals are for exact copied work without any additions at all. \nPerhaps even more people leveraged late solutions in their blends. This system promotes people to think less, not more. I wonder how many of these folks even understand what they copy - we don't have a way to tell.\n\nI wonder if people who submit these solutions really want to showcase such a result on resume... This is such a shame show. To discourage it, I could suggest Kaggle to \n\n* only give credit to creator of the first published solution among identical submissions (even if it is not selected as final, might want to think a bit more of what we call identical); we should not allow identical solution to be selected for private leaderboard submission for people who copy it.\n* publish a few more decimals of the private score and highlight identical submission on leaderboard as duplicate, to make it even more obvious that it's exact same solutions being submitted\n\nI would fully agree for restricted sharing in the end of the competition as well. Typically people focus on working on their solutions at this time, not everyone has a chance to react to public kernel in a short period of time even if they wanted to. \n\nI think also it makes sense for Kaggle to retro-actively adjust results of this competition itself to help recognize the modeling work that people put in. \n\nHope this helps figure out a good approach that promotes excellence and collaboration, while giving credit where it's due.", "324416": "Why would recruiters use kaggle now that these blinking kernels exist. Kaggle sort it out - you will lose hard working competitors too! When I wrote the kernels schmernels post I was inaccurate when I said start two weeks from the end - apologies ;(", "325412": "One thing to not forget - if ***that kernel*** hadn't been released - this was a very very successful competition overall - lots of scope for feature engineering that didn't have leakage!", "324393": "It's very disappointing to see LB being destroyed in last hours. I am sure that Kaggle can't do anything about it because they will never know who will be posting what and when. But at least they can have some mechanism to allow community to punish such `D***s` flying around in every competition in last hours. \n\nThis is regular problem and needs a solution. I and many others strongly recommend having **downvote option for kernels** . \n\nMany kagglers have already educated that novice kernel author but it seems that he is purely in the mood of destroying leaderboard and not taking down that kernel. So may be there is a need to have a Hall of Shame post or a kernel like [this][1] \n\n\n [1]: https://www.kaggle.com/carloshuertas/kaggle-users-with-most-awful-forum-karma", "324358": "within last hour, many with <5 subs jumps to this score. Now imagine you were someone having 9810 last night and thought you could get your first silver medal.", "325047": "as a new player of kaggle, I just want to cry. Very disappointed.", "325027": "Very bad Kaggle experience this morning.", "324411": "It is interesting to see how many people are willing to use the infamous late kernel to obtain a now meaningless high rank in this competition. Maybe that should be the next competition: predict which Kagglers will make the late submission of someone else's work.", "324384": "It is worse now there are datasets - I have been warning about this for months - you just get a \"we can't do anything about it\" which is something I would fire someone over if they worked for me (I am a bad boss I suppose)", "324425": "I think, that this topic has made 0.9811 kernel much more popular.", "324405": "I went to sleep in silver medal position, then I woke up and found that infamous kernel. I have made a significant effort and I was enjoying the competition, now this situation is very frustrating. Now ,as many competitors, I am at work and don't have the time and resources to train more models. So, kaggle user who want to use this kernel consider the fairness of the competition and don't use it please. ", "324345": "Agree with you, investing big time and the last day you find high score Kernal where no time to learn from it or maybe you don't have submissions left to use it ! then hours of your afford vanish.\n\n**My suggestion after the rules acceptance deadline to stop public Kernals because no much time to learn from it. OR at least stop upload input to public Kernal at that date**\n\nAlso regarding Kaggle removing scores of competitors because of cheating usually happen for unaware of rules to publish results primitively or share it with friend, I suggest Kaggle make a run to remove them at acceptance deadline, As people decide to invest more and more time on competition after it or could decide to rent a server and pay money then find score vanished by the end. Also more fair to have time to send appeal in case required.", "324767": "I want to resign from this competition.\n\nYes, the ranks matter (it is a competition!) and yes I m pissed that my participation is going to benefit last minute public blending competitors!", "324991": "Is it cheating or unethical? I am completely against the term **cheating** here. There's no rule in the competition that describes public sharing at the last moment is not allowed and will be termed as \"Cheater\"\n\nCouple of cases -\n> Somebody uses the csv to upload. -- Is it against rule?\n> Somebody blended with their model-- is it cheating?\n> Somebody uses some of their features and used it ( like me) in their model. -- what would you call it?\n> Somebody who did their blending at last moment with/without using this kernel's CSV. -- can it be tracked?\n\nSomehow I would term the entire episode as Unethical rather than cheating. I am not sure what Kaggle is going to do today but its unfair to remove all those who directly/indirectly uses the kernel. Rather as already mentioned , Kaggle should do something in future competition to stop the practice.", "325334": "I would say no sharing of public kernels that can achieve top 20% score. It was disappointing I was in the top 6% last night. Got up to find 300 new people on top of me with the same score!! ARGH!!", "324655": "How about we post a 0.9832 solution? (but with random noise in the private hours) \nJust kidding...", "324782": "There were talks about blends spoiling the community, blending attempts at last minute of this competition has made this even worse.", "324559": "For all those punters wishing to sleep tonight click on the option to unsubscribe ;)", "324522": "Can I withdraw from the competition?\n\nI already have two dark spots on my portfolio:\n\nSanta Gift Matching Challenge - Infinite Probabilistic Improver was published when I was on holiday\n\nMercedes-Benz Greener Manufacturing - for most of us it seems the [Manatees][1] decided the place on LB\n\nNow this.\n\n\n[1]: http://southpark.wikia.com/wiki/Manatees", "324456": "It seems there are already a lot of good suggestions here. My question is how to get Kaggle officials' attention and act on it. Personally I think it is bad for Kaggle as a company as well for various reasons:\n\n1) The most attractive part of Kaggle is the sheer amount of excellent data scientists that are actively learning, sharing and developing methods. Last-minute high-score kernel/results sharing will definitely disappoint and drive away many good and hard working contributors, which I don't think Kaggle wanna see. \n\n2) My guesses are part of Kaggle's business model is for companies to recruit from Kaggle platform. If things like this happen, where the ranking of a competition become rather meaningless, I don't think recruiter will see any value here. As a matter of fact, I found more and more recruiters and people in data science community think Kaggle is just for competition but not real data science experience/projects. I think that is bad for the company as well. \n\nIt takes a long time and effort to build a platform this great and have so many wonderful data scientists (I am relatively new, but already learnt a lot from many discussions posts and kernels). I hope there is a way to keep improving it, keep all the great people around and keep the competition rankings more meaningful. ", "324233": "If people can submit exact same results, even from public kernels, I think it is weird and does not help the ecosystem and should be considered as cunning. \nhttps://www.kaggle.com/general/48852", "324894": "Was really disappointed to see the blends not be punished for overfitting. I guess leaderboard is kind of a validation set :/\n\nMy final model: \nPublic AUC .9812 Private AUC .9822\nBlend:\nPublic AUC .9812 Private AUC .9820", "324401": "I am new and naive on this subject, but would it be possible or feasible to handle this issue the way patents on IP (intellectual property) are handled? So many products have multiple patent licensing agreements among many patent holders. Obviously if the administrative overhead would be too much, it would not be worth it. Or if things go the way of patent trolls and endless \"litigation\" that's no good either. If there is a way to give credit for an original idea to the originator, that makes sense to me and seems fair. ", "325136": "As a beginner, I think there should be a closed period for games near the end.", "324245": "First of all, this is a great post. I agree with most of the things you mentioned here. \n\nI will tell you about my experience. I joined like 5 months ago and was unable to approach any competition other than Titanic. For some time, I questioned my decision of joining kaggle. I thought it was too early. That time kernels came to my rescue. I have learned something new while reading kernels every day. I would say there has been a great improvement in my performance as a data scientist. Kernels have surely changed the kaggle ecosystem. Now, people can learn from others' work. There is so much content at one place that you don't have to go anywhere else to increase your knowledge.\n\nBut on the other side, kernels affect competitions by increasing the ranks of people who just fork and submit. If this has resulted in older users straying away from the platform, it is more serious than I thought. To solve this, I would recommend something like providing proof of work while submitting for competitions. Everyone should also upload their scripts used for the competition. Now, if they prepare kernels on kaggle itself and wish to make them public, they should be able to do that only after the competition is over. \n\nI don't agree with your suggestion of introducing downvotes for kernels. It takes significant amount of time and hard work to create a kernel. If one can't praise someone's hard work, he/she should also not degrade it by providing downvotes. Downvotes can bring about a negative ecosystem on this platform. ", "325414": "Maybe we need a few *pre-checkings* before metric calculation? Like [numerai's scoring part][1].\n> ...*Copied from https://numer.ai/learn*\n> \n> **Consistency** measures the percentage of eras in which a model achieves a logloss better than the benchmark. Numerai wants models that work well consistently across eras. Only models with consistency above 58% are considered consistent.\n> \n> **Originality** is a measure of whether a set of predictions is uncorrelated with predictions already submitted. Numerai wants to encourage new models over duplicate submissions.\n> \n> **Concordance** is a measure of whether predictions on the validation set, test set, and live set appread to be generated by the same model. A data scientist who submits perfect answers on the validation set is unlikely to achieve concordance.\n\nWhile it may be quite difficult to design perfect checking functions, **Originality** should be taken care, at least...\n [1]: https://numer.ai/learn", "324529": "I'm new and obviously have no experience with such events. From what I can tell it would be impossible for Kaggle to identify who used \"the kernel\" only by using the submission. Obviously it's easy if its a perfect match but not so easy if any kind of blending has been done. \n\nOne solution I see for people that deserve medals to get their hard work properly rewarded would be to ask people to submit the model that generates the solution. This way it should be a bit simpler to identify and disqualify users of \"the kernel\". If no model is uploaded the medal should not be awarded. \n\nWhat do you guys think?", "325426": "+1", "325578": "I consider Kaggle as a educational platform, and the kernels are best materials to learn data science.\nI think it is impossible to ban the kernels in any forms, because it violets the spirit of Kaggle.\nKernels are public resources and every participant share the results, which helps everyone to improve.\nHowever, why not AT LEAST, AT LEAST, AT LEAST blend the results on our own.\n\nIn short:\nMain problem = Public Blending Kernels", "324458": "Generally, I find high-scoring public LB models to actually be pretty helpful as I can blend them with my own and improve my own score. Now, I totally get the frustration involved in the hundreds of blending posts - they should be used to serve a purpose such as demonstrate that highly diverse low-scoring models can be blended to perform far better, or how you can determine weights for your ensembling model.", "324771": "I understand that dropping 100 positions within a day or within a few hours can be really frustrating. But does it really matter in the end? If you finish #200 or #300? If you get bronze or not? In the beginning it is only the learning process that matters. I personally learned a lot from kernels and improved my skills ever since I started kaggeling. I am very thankful that Kaggle is providing this great platform. So no need to complain, Kaggle is giving their best, especially for starters.\n\nPS: And if you care about the rank, take the public kernel, blend it with your own model and there you have a jump on LB. That's what I did in my first competition and it worked pretty well. Of course I don't like to drop 100 positions as well :)", "324760": "This kind of kernels can be very frustrating, but I would argue that ranks, medals and points don't have much value themselves. It is knowledge behind them that matters.", "327851": "How about we keep the kernel private and adding the share option for private kernel, that sound ridiculous but that might keep the reputation. ", "325660": "Kaggle is probably doing something for it.. who knows? \nThey are yet to finalise the final standing..", "324490": "It's a pity seeing that competitors like us are using that infamous kernel.", "343430": "", "326171": "", "325574": "", "325451": "", "324267": "" } }