Hi everybody,
\n\nI am a newbie in datascience which I just didscovered 2 years ago.\nUnfortunately, I am an old french man and my english writing is not so good ;-)
\n\nAnyway, I am not sure to understand the purpose of public kernels in these challenges.\nYes, it helps beginners, I confirm, but it seems to skew the result on the public leaderboard.
\n\nI am wondering which it is the need to publish these scores on the public leaderboard ?
\n\nThank you for helping.
\n\nGilles.
", "messages": [ { "id": "534177", "postDate": "05/20/2019 21:15:33", "content": "Hi everybody,
\n\nI am a newbie in datascience which I just didscovered 2 years ago.\nUnfortunately, I am an old french man and my english writing is not so good ;-)
\n\nAnyway, I am not sure to understand the purpose of public kernels in these challenges.\nYes, it helps beginners, I confirm, but it seems to skew the result on the public leaderboard.
\n\nI am wondering which it is the need to publish these scores on the public leaderboard ?
\n\nThank you for helping.
\n\nGilles.
", "rawMarkdown": "Hi everybody,\n\nI am a newbie in datascience which I just didscovered 2 years ago.\nUnfortunately, I am an old french man and my english writing is not so good ;-)\n\nAnyway, I am not sure to understand the purpose of public kernels in these challenges.\nYes, it helps beginners, I confirm, but it seems to skew the result on the public leaderboard.\n\nI am wondering which it is the need to publish these scores on the public leaderboard ?\n\nThank you for helping.\n\nGilles.", "votes": null }, { "id": "534190", "postDate": "05/20/2019 22:03:37", "content": "The public kernels have been controversial for a while in Kaggle, precisely for the reason that you mention. \nSee this post for example:\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182
\n\nHowever, recently the situation has stabilized and Kagglers seem to have understood that publishing high scoring kernels should be avoided towards the end of the competition. I think the kernel tab will show this kind of warning too, in a few days.\nI am considering \"high-scoring\" whatever has the potential to enter in the medal zone as a single kernel.
", "rawMarkdown": "The public kernels have been controversial for a while in Kaggle, precisely for the reason that you mention. \nSee this post for example:\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182\n\nHowever, recently the situation has stabilized and Kagglers seem to have understood that publishing high scoring kernels should be avoided towards the end of the competition. I think the kernel tab will show this kind of warning too, in a few days.\nI am considering \"high-scoring\" whatever has the potential to enter in the medal zone as a single kernel.", "votes": null }, { "id": "534456", "postDate": "05/21/2019 09:16:46", "content": "Unfortunately to get a gold medal in Kaggle Kernels - 1.) An EDA early in the competition or 2.) A high scoring kernel later on in the competition.
\n\nLuckily most high scoring kernels crash and burn on the private LB so don't worry too much! ;)\nWe have only 16 earthquakes to train so couple with MAE the variance will be huge - this could be the biggest shakeup ever!
", "rawMarkdown": "Unfortunately to get a gold medal in Kaggle Kernels - 1.) An EDA early in the competition or 2.) A high scoring kernel later on in the competition.\n\nLuckily most high scoring kernels crash and burn on the private LB so don't worry too much! ;)\nWe have only 16 earthquakes to train so couple with MAE the variance will be huge - this could be the biggest shakeup ever!", "votes": null }, { "id": "534459", "postDate": "05/21/2019 09:21:11", "content": "LOL, I used to share EDA kernels late in the competition.. Now I know why I am low in kernel ranking ;)
", "rawMarkdown": "LOL, I used to share EDA kernels late in the competition.. Now I know why I am low in kernel ranking ;)", "votes": null }, { "id": "534462", "postDate": "05/21/2019 09:25:38", "content": "Well, I have 4 gold kernels in this competition and not all of them were written early or have high score :)
", "rawMarkdown": "Well, I have 4 gold kernels in this competition and not all of them were written early or have high score :)", "votes": null }, { "id": "534487", "postDate": "05/21/2019 10:13:54", "content": "@artgor - your worst scoring kernel was the best scoring kernel for quite a while - that is why I stole your features ;) The other three are in the top 10!!
", "rawMarkdown": "artgor - your worst scoring kernel was the best scoring kernel for quite a while - that is why I stole your features ;) The other three are in the top 10!!", "votes": null }, { "id": "534550", "postDate": "05/21/2019 12:57:21", "content": "The competition hosts don't care who wins or who gets a medal. They just want the best possible solution for their problem. Public kernels push the boundaries of the solutions and that is desirable by the host. Kaggle gets paid by the hosts, so why would they want to force any rule regarding this! \nIt's about business, not about fair play.
", "rawMarkdown": "The competition hosts don't care who wins or who gets a medal. They just want the best possible solution for their problem. Public kernels push the boundaries of the solutions and that is desirable by the host. Kaggle gets paid by the hosts, so why would they want to force any rule regarding this! \nIt's about business, not about fair play.", "votes": null }, { "id": "534556", "postDate": "05/21/2019 13:04:39", "content": "I honestly don't think that is the reason for introducing the public kernels, but I may be wrong...\nPersonally I found them extremely useful to tackle data that was out of my domain, like images not long ago. I also use them as inspiration in generating my features, like I did a lot in this competition.
", "rawMarkdown": "I honestly don't think that is the reason for introducing the public kernels, but I may be wrong...\nPersonally I found them extremely useful to tackle data that was out of my domain, like images not long ago. I also use them as inspiration in generating my features, like I did a lot in this competition.", "votes": null }, { "id": "534573", "postDate": "05/21/2019 13:33:34", "content": "I agree kernels are great resources. I learned a lot from kernels in this competition and in closed competitions. One limitation for me was that I am using R, so basically I had nothing to benefit from public kernels. \nBut I guess OP is talking here about high-scoring kernels. The good example in this competition is the genetic programming kernels. They are completely black boxes, nothing to be learnt from. Their only job is to give a higher position on the LB (no offense to Scirpus).
", "rawMarkdown": "I agree kernels are great resources. I learned a lot from kernels in this competition and in closed competitions. One limitation for me was that I am using R, so basically I had nothing to benefit from public kernels. \nBut I guess OP is talking here about high-scoring kernels. The good example in this competition is the genetic programming kernels. They are completely black boxes, nothing to be learnt from. Their only job is to give a higher position on the LB (no offense to Scirpus).", "votes": null }, { "id": "534656", "postDate": "05/21/2019 16:06:47", "content": "I found a discussion that presents similar questions as you. It may be helpful, but it may be a philosophical good question:)\nhttps://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/51561#293906
\n\nI think that public kernels are creating an opportunity to emerge, such as building better models and creating new ideas, as people are more focused on the public kernel during the competition. Of course I don't think it's just that good side.
", "rawMarkdown": "I found a discussion that presents similar questions as you. It may be helpful, but it may be a philosophical good question:)\nhttps://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/51561#293906\n\nI think that public kernels are creating an opportunity to emerge, such as building better models and creating new ideas, as people are more focused on the public kernel during the competition. Of course I don't think it's just that good side.", "votes": null }, { "id": "534784", "postDate": "05/21/2019 20:25:30", "content": "I think some people like the feeling of having posted the highest scoring public kernel. What a pity this is impossible due to one of the first kernels! https://www.kaggle.com/inversion/basic-feature-benchmark
", "rawMarkdown": "I think some people like the feeling of having posted the highest scoring public kernel. What a pity this is impossible due to one of the first kernels! https://www.kaggle.com/inversion/basic-feature-benchmark", "votes": null }, { "id": "534797", "postDate": "05/21/2019 20:53:32", "content": "public kernels are the reason i switched from R to Python
", "rawMarkdown": "public kernels are the reason i switched from R to Python", "votes": null }, { "id": "534809", "postDate": "05/21/2019 21:33:37", "content": "I never thought about that! We definitely have less R scripts these days.
", "rawMarkdown": "I never thought about that! We definitely have less R scripts these days.", "votes": null }, { "id": "535264", "postDate": "05/22/2019 15:27:46", "content": "Maybe they wanna get medals on kernels
", "rawMarkdown": "Maybe they wanna get medals on kernels", "votes": null }, { "id": "535624", "postDate": "05/23/2019 08:32:14", "content": "I've also switched form R to python, however for an R user translating python kernels is a good exercise to learn
", "rawMarkdown": "I've also switched form R to python, however for an R user translating python kernels is a good exercise to learn", "votes": null }, { "id": "535673", "postDate": "05/23/2019 10:19:27", "content": "\n\n\nIt's about business, not about fair play.
\n
It is Kaggle long term interested to keep a set of high quality and motivated competitors so that it can sustain its business. Fairness is key for that. Indeed, competitors come to learn and get some recognition of what they have learned. Few come for the money given it is very unlikely to get prize money. If you end up down in the LB because someone shared a high quality kernel late in the competition and you don't have time to move above it then you're pissed off. This happened in the past, and I hope it won't happen again.
", "rawMarkdown": "> It's about business, not about fair play.\n\nIt is Kaggle long term interested to keep a set of high quality and motivated competitors so that it can sustain its business. Fairness is key for that. Indeed, competitors come to learn and get some recognition of what they have learned. Few come for the money given it is very unlikely to get prize money. If you end up down in the LB because someone shared a high quality kernel late in the competition and you don't have time to move above it then you're pissed off. This happened in the past, and I hope it won't happen again.", "votes": null }, { "id": "535680", "postDate": "05/23/2019 10:41:13", "content": "I have only two public kernels on kaggle. Total upvotes = 3 out of which 2 are mine 😆 at least I kept myself motivated. Thanks for sharing kernel strategy.
", "rawMarkdown": "I have only two public kernels on kaggle. Total upvotes = 3 out of which 2 are mine 😆 at least I kept myself motivated. Thanks for sharing kernel strategy.", "votes": null }, { "id": "535685", "postDate": "05/23/2019 10:48:18", "content": "\n\n\npublic kernels are the reason i switched from R to Python
\n
I feel world would be a better place if R gets the attention it deserves. I switched to python because of ease to pair up with teams in industry as there are lot more python users as there are R users. But still have a huge crush on R. There are some out of the world packages such as LaplaceDemon and causal inference etc etc.
", "rawMarkdown": "> public kernels are the reason i switched from R to Python\n\n**I feel world would be a better place if R gets the attention it deserves.** I switched to python because of ease to pair up with teams in industry as there are lot more python users as there are R users. But still have a huge crush on R. There are some out of the world packages such as LaplaceDemon and causal inference etc etc.", "votes": null }, { "id": "535803", "postDate": "05/23/2019 13:21:34", "content": "\n\n\nBut still have a huge crush on R
\n
I just really miss data.table and how easy plotting is in R. I know H2O.ai is working on Python datatable but it's not on Kaggle yet, so I can't use it. There's also just soooo many other packages. I like mice imputation; I also like functions such as paste
\n\n\nbut it's not on Kaggle yet, so I can't use it
\n
@returnofsputnik You can install python data.table by connecting kernel to internet, I have used it on kaggle works fine. Installing from pre made wheel is recommended look this by two GM who work at h2o.ai https://www.kaggle.com/tunguz/eda-with-python-datatable#465526 and another one https://www.kaggle.com/sudalairajkumar/getting-started-with-python-datatable
", "rawMarkdown": "> but it's not on Kaggle yet, so I can't use it\n\n@returnofsputnik You can install python data.table by connecting kernel to internet, I have used it on kaggle works fine. Installing from pre made wheel is recommended look this by two GM who work at h2o.ai https://www.kaggle.com/tunguz/eda-with-python-datatable#465526 and another one https://www.kaggle.com/sudalairajkumar/getting-started-with-python-datatable", "votes": null }, { "id": "536053", "postDate": "05/23/2019 21:32:22", "content": "Hello bluetrain,\nThank you for your reply.\nAs I mentionned in my initial post, my english is not so good and unfortunately I can't reply to every post because of lack of time.\nThank you to everybody.
\n\nOn my own, I am a beginner and I agree, kernels help to learn a lot.\nHowever, it can also discourage many kagglers who refuse by ethics maybe to fork kernels and blend it just to go up in the public leaderboard.
\n\nAnyway, congratulations to you and your ABC team for your new score.
\n\nIt is amazing for me ;-)
\n\nGilles.
", "rawMarkdown": "Hello bluetrain,\nThank you for your reply.\nAs I mentionned in my initial post, my english is not so good and unfortunately I can't reply to every post because of lack of time.\nThank you to everybody.\n\nOn my own, I am a beginner and I agree, kernels help to learn a lot.\nHowever, it can also discourage many kagglers who refuse by ethics maybe to fork kernels and blend it just to go up in the public leaderboard.\n \nAnyway, congratulations to you and your ABC team for your new score.\n\nIt is amazing for me ;-)\n\nGilles.", "votes": null }, { "id": "536300", "postDate": "05/24/2019 08:06:53", "content": "@gillou993 \nFirst of all: I think your English is good and you should not be ashamed about it ;)
\n\nI must say, at the begin I had your same feeling about public kernels. I was wondering why a competition could not just be a Competition, with everyone working his best on his own (team). \nBut after a while I started appreciating kernels, as they allow me to get started in domains where I had no clue where to start from. I could grasp interesting ideas and try to improve them. I could improve my visualization skills (I am still not good at that though) etc. etc.\nI hope you will be able to get the good of public kernels too at some point.
\n\nThanks for your kind words. Wish you the best for the rest of the competition!
", "rawMarkdown": "gillou993 \nFirst of all: I think your English is good and you should not be ashamed about it ;)\n\nI must say, at the begin I had your same feeling about public kernels. I was wondering why a competition could not just be a Competition, with everyone working his best on his own (team). \nBut after a while I started appreciating kernels, as they allow me to get started in domains where I had no clue where to start from. I could grasp interesting ideas and try to improve them. I could improve my visualization skills (I am still not good at that though) etc. etc.\nI hope you will be able to get the good of public kernels too at some point.\n\nThanks for your kind words. Wish you the best for the rest of the competition!", "votes": null }, { "id": "536628", "postDate": "05/24/2019 21:23:56", "content": "Hello,\nOk thank you ;-)
\n\nActually, I can read english relatively easely , but it is much more complicated for me writing.\nIt is a bit frustrating because I would like to participate in discussions but it takes me too much time\nto write what I am thinking precisely...
\n\nTo get back to the initial subject of this post, I agree, kernels can be a great help for beginners.\nNevertheless, public kernels may also drive beginners to the darkside of ML :-)
\n\nI see a lot of kernels with hundreds or thousands of random generated features in this competition, I am a newbie but I am surprised with this gait. I am not convinced it is the right way.
\n\nWish you the best for the rest of the competition too !
\n\nGilles.
", "rawMarkdown": "Hello,\nOk thank you ;-)\n\nActually, I can read english relatively easely , but it is much more complicated for me writing.\nIt is a bit frustrating because I would like to participate in discussions but it takes me too much time\nto write what I am thinking precisely...\n\nTo get back to the initial subject of this post, I agree, kernels can be a great help for beginners.\nNevertheless, public kernels may also drive beginners to the darkside of ML :-)\n\nI see a lot of kernels with hundreds or thousands of random generated features in this competition, I am a newbie but I am surprised with this gait. I am not convinced it is the right way.\n\nWish you the best for the rest of the competition too !\n\nGilles.", "votes": null } ], "comments": [ { "id": 534190, "author_name": "stecasasso", "author_url": "", "post_date": "05/20/2019 22:03:37", "content": "The public kernels have been controversial for a while in Kaggle, precisely for the reason that you mention. \nSee this post for example:\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182
\n\nHowever, recently the situation has stabilized and Kagglers seem to have understood that publishing high scoring kernels should be avoided towards the end of the competition. I think the kernel tab will show this kind of warning too, in a few days.\nI am considering \"high-scoring\" whatever has the potential to enter in the medal zone as a single kernel.
", "votes": null, "replies": [ { "id": 536053, "author_name": "gillou993", "author_url": "", "post_date": "05/23/2019 21:32:22", "content": "Hello bluetrain,\nThank you for your reply.\nAs I mentionned in my initial post, my english is not so good and unfortunately I can't reply to every post because of lack of time.\nThank you to everybody.
\n\nOn my own, I am a beginner and I agree, kernels help to learn a lot.\nHowever, it can also discourage many kagglers who refuse by ethics maybe to fork kernels and blend it just to go up in the public leaderboard.
\n\nAnyway, congratulations to you and your ABC team for your new score.
\n\nIt is amazing for me ;-)
\n\nGilles.
", "votes": null, "replies": [] }, { "id": 536300, "author_name": "stecasasso", "author_url": "", "post_date": "05/24/2019 08:06:53", "content": "@gillou993 \nFirst of all: I think your English is good and you should not be ashamed about it ;)
\n\nI must say, at the begin I had your same feeling about public kernels. I was wondering why a competition could not just be a Competition, with everyone working his best on his own (team). \nBut after a while I started appreciating kernels, as they allow me to get started in domains where I had no clue where to start from. I could grasp interesting ideas and try to improve them. I could improve my visualization skills (I am still not good at that though) etc. etc.\nI hope you will be able to get the good of public kernels too at some point.
\n\nThanks for your kind words. Wish you the best for the rest of the competition!
", "votes": null, "replies": [] }, { "id": 536628, "author_name": "gillou993", "author_url": "", "post_date": "05/24/2019 21:23:56", "content": "Hello,\nOk thank you ;-)
\n\nActually, I can read english relatively easely , but it is much more complicated for me writing.\nIt is a bit frustrating because I would like to participate in discussions but it takes me too much time\nto write what I am thinking precisely...
\n\nTo get back to the initial subject of this post, I agree, kernels can be a great help for beginners.\nNevertheless, public kernels may also drive beginners to the darkside of ML :-)
\n\nI see a lot of kernels with hundreds or thousands of random generated features in this competition, I am a newbie but I am surprised with this gait. I am not convinced it is the right way.
\n\nWish you the best for the rest of the competition too !
\n\nGilles.
", "votes": null, "replies": [] } ] }, { "id": 534456, "author_name": "scirpus", "author_url": "", "post_date": "05/21/2019 09:16:46", "content": "Unfortunately to get a gold medal in Kaggle Kernels - 1.) An EDA early in the competition or 2.) A high scoring kernel later on in the competition.
\n\nLuckily most high scoring kernels crash and burn on the private LB so don't worry too much! ;)\nWe have only 16 earthquakes to train so couple with MAE the variance will be huge - this could be the biggest shakeup ever!
", "votes": null, "replies": [ { "id": 534459, "author_name": "cpmpml", "author_url": "", "post_date": "05/21/2019 09:21:11", "content": "LOL, I used to share EDA kernels late in the competition.. Now I know why I am low in kernel ranking ;)
", "votes": null, "replies": [] }, { "id": 534462, "author_name": "artgor", "author_url": "", "post_date": "05/21/2019 09:25:38", "content": "Well, I have 4 gold kernels in this competition and not all of them were written early or have high score :)
", "votes": null, "replies": [] }, { "id": 534487, "author_name": "scirpus", "author_url": "", "post_date": "05/21/2019 10:13:54", "content": "@artgor - your worst scoring kernel was the best scoring kernel for quite a while - that is why I stole your features ;) The other three are in the top 10!!
", "votes": null, "replies": [] }, { "id": 534784, "author_name": "bigironsphere", "author_url": "", "post_date": "05/21/2019 20:25:30", "content": "I think some people like the feeling of having posted the highest scoring public kernel. What a pity this is impossible due to one of the first kernels! https://www.kaggle.com/inversion/basic-feature-benchmark
", "votes": null, "replies": [] }, { "id": 535680, "author_name": "cyberia", "author_url": "", "post_date": "05/23/2019 10:41:13", "content": "I have only two public kernels on kaggle. Total upvotes = 3 out of which 2 are mine 😆 at least I kept myself motivated. Thanks for sharing kernel strategy.
", "votes": null, "replies": [] } ] }, { "id": 534550, "author_name": "amjad85", "author_url": "", "post_date": "05/21/2019 12:57:21", "content": "The competition hosts don't care who wins or who gets a medal. They just want the best possible solution for their problem. Public kernels push the boundaries of the solutions and that is desirable by the host. Kaggle gets paid by the hosts, so why would they want to force any rule regarding this! \nIt's about business, not about fair play.
", "votes": null, "replies": [ { "id": 534556, "author_name": "stecasasso", "author_url": "", "post_date": "05/21/2019 13:04:39", "content": "I honestly don't think that is the reason for introducing the public kernels, but I may be wrong...\nPersonally I found them extremely useful to tackle data that was out of my domain, like images not long ago. I also use them as inspiration in generating my features, like I did a lot in this competition.
", "votes": null, "replies": [] }, { "id": 534573, "author_name": "amjad85", "author_url": "", "post_date": "05/21/2019 13:33:34", "content": "I agree kernels are great resources. I learned a lot from kernels in this competition and in closed competitions. One limitation for me was that I am using R, so basically I had nothing to benefit from public kernels. \nBut I guess OP is talking here about high-scoring kernels. The good example in this competition is the genetic programming kernels. They are completely black boxes, nothing to be learnt from. Their only job is to give a higher position on the LB (no offense to Scirpus).
", "votes": null, "replies": [] }, { "id": 534797, "author_name": "returnofsputnik", "author_url": "", "post_date": "05/21/2019 20:53:32", "content": "public kernels are the reason i switched from R to Python
", "votes": null, "replies": [] }, { "id": 534809, "author_name": "scirpus", "author_url": "", "post_date": "05/21/2019 21:33:37", "content": "I never thought about that! We definitely have less R scripts these days.
", "votes": null, "replies": [] }, { "id": 535624, "author_name": "tfrmarques", "author_url": "", "post_date": "05/23/2019 08:32:14", "content": "I've also switched form R to python, however for an R user translating python kernels is a good exercise to learn
", "votes": null, "replies": [] }, { "id": 535673, "author_name": "cpmpml", "author_url": "", "post_date": "05/23/2019 10:19:27", "content": "\n\n\nIt's about business, not about fair play.
\n
It is Kaggle long term interested to keep a set of high quality and motivated competitors so that it can sustain its business. Fairness is key for that. Indeed, competitors come to learn and get some recognition of what they have learned. Few come for the money given it is very unlikely to get prize money. If you end up down in the LB because someone shared a high quality kernel late in the competition and you don't have time to move above it then you're pissed off. This happened in the past, and I hope it won't happen again.
", "votes": null, "replies": [] }, { "id": 535685, "author_name": "cyberia", "author_url": "", "post_date": "05/23/2019 10:48:18", "content": "\n\n\npublic kernels are the reason i switched from R to Python
\n
I feel world would be a better place if R gets the attention it deserves. I switched to python because of ease to pair up with teams in industry as there are lot more python users as there are R users. But still have a huge crush on R. There are some out of the world packages such as LaplaceDemon and causal inference etc etc.
", "votes": null, "replies": [] }, { "id": 535803, "author_name": "returnofsputnik", "author_url": "", "post_date": "05/23/2019 13:21:34", "content": "\n\n\nBut still have a huge crush on R
\n
I just really miss data.table and how easy plotting is in R. I know H2O.ai is working on Python datatable but it's not on Kaggle yet, so I can't use it. There's also just soooo many other packages. I like mice imputation; I also like functions such as paste
\n\n\nbut it's not on Kaggle yet, so I can't use it
\n
@returnofsputnik You can install python data.table by connecting kernel to internet, I have used it on kaggle works fine. Installing from pre made wheel is recommended look this by two GM who work at h2o.ai https://www.kaggle.com/tunguz/eda-with-python-datatable#465526 and another one https://www.kaggle.com/sudalairajkumar/getting-started-with-python-datatable
", "votes": null, "replies": [] } ] }, { "id": 534656, "author_name": "dhaqui", "author_url": "", "post_date": "05/21/2019 16:06:47", "content": "I found a discussion that presents similar questions as you. It may be helpful, but it may be a philosophical good question:)\nhttps://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/51561#293906
\n\nI think that public kernels are creating an opportunity to emerge, such as building better models and creating new ideas, as people are more focused on the public kernel during the competition. Of course I don't think it's just that good side.
", "votes": null, "replies": [] }, { "id": 535264, "author_name": "", "author_url": "", "post_date": "05/22/2019 15:27:46", "content": "Maybe they wanna get medals on kernels
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "534177": "Hi everybody,\n\nI am a newbie in datascience which I just didscovered 2 years ago.\nUnfortunately, I am an old french man and my english writing is not so good ;-)\n\nAnyway, I am not sure to understand the purpose of public kernels in these challenges.\nYes, it helps beginners, I confirm, but it seems to skew the result on the public leaderboard.\n\nI am wondering which it is the need to publish these scores on the public leaderboard ?\n\nThank you for helping.\n\nGilles.", "534190": "The public kernels have been controversial for a while in Kaggle, precisely for the reason that you mention. \nSee this post for example:\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182\n\nHowever, recently the situation has stabilized and Kagglers seem to have understood that publishing high scoring kernels should be avoided towards the end of the competition. I think the kernel tab will show this kind of warning too, in a few days.\nI am considering \"high-scoring\" whatever has the potential to enter in the medal zone as a single kernel.", "534456": "Unfortunately to get a gold medal in Kaggle Kernels - 1.) An EDA early in the competition or 2.) A high scoring kernel later on in the competition.\n\nLuckily most high scoring kernels crash and burn on the private LB so don't worry too much! ;)\nWe have only 16 earthquakes to train so couple with MAE the variance will be huge - this could be the biggest shakeup ever!", "534459": "LOL, I used to share EDA kernels late in the competition.. Now I know why I am low in kernel ranking ;)", "534462": "Well, I have 4 gold kernels in this competition and not all of them were written early or have high score :)", "534487": "artgor - your worst scoring kernel was the best scoring kernel for quite a while - that is why I stole your features ;) The other three are in the top 10!!", "534550": "The competition hosts don't care who wins or who gets a medal. They just want the best possible solution for their problem. Public kernels push the boundaries of the solutions and that is desirable by the host. Kaggle gets paid by the hosts, so why would they want to force any rule regarding this! \nIt's about business, not about fair play.", "534556": "I honestly don't think that is the reason for introducing the public kernels, but I may be wrong...\nPersonally I found them extremely useful to tackle data that was out of my domain, like images not long ago. I also use them as inspiration in generating my features, like I did a lot in this competition.", "534573": "I agree kernels are great resources. I learned a lot from kernels in this competition and in closed competitions. One limitation for me was that I am using R, so basically I had nothing to benefit from public kernels. \nBut I guess OP is talking here about high-scoring kernels. The good example in this competition is the genetic programming kernels. They are completely black boxes, nothing to be learnt from. Their only job is to give a higher position on the LB (no offense to Scirpus).", "534656": "I found a discussion that presents similar questions as you. It may be helpful, but it may be a philosophical good question:)\nhttps://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/51561#293906\n\nI think that public kernels are creating an opportunity to emerge, such as building better models and creating new ideas, as people are more focused on the public kernel during the competition. Of course I don't think it's just that good side.", "534784": "I think some people like the feeling of having posted the highest scoring public kernel. What a pity this is impossible due to one of the first kernels! https://www.kaggle.com/inversion/basic-feature-benchmark", "534797": "public kernels are the reason i switched from R to Python", "534809": "I never thought about that! We definitely have less R scripts these days.", "535264": "Maybe they wanna get medals on kernels", "535624": "I've also switched form R to python, however for an R user translating python kernels is a good exercise to learn", "535673": "> It's about business, not about fair play.\n\nIt is Kaggle long term interested to keep a set of high quality and motivated competitors so that it can sustain its business. Fairness is key for that. Indeed, competitors come to learn and get some recognition of what they have learned. Few come for the money given it is very unlikely to get prize money. If you end up down in the LB because someone shared a high quality kernel late in the competition and you don't have time to move above it then you're pissed off. This happened in the past, and I hope it won't happen again.", "535680": "I have only two public kernels on kaggle. Total upvotes = 3 out of which 2 are mine 😆 at least I kept myself motivated. Thanks for sharing kernel strategy.", "535685": "> public kernels are the reason i switched from R to Python\n\n**I feel world would be a better place if R gets the attention it deserves.** I switched to python because of ease to pair up with teams in industry as there are lot more python users as there are R users. But still have a huge crush on R. There are some out of the world packages such as LaplaceDemon and causal inference etc etc.", "535803": "> But still have a huge crush on R\n\nI just really miss `data.table` and how easy plotting is in R. I know H2O.ai is working on Python `datatable` but it's not on Kaggle yet, so I can't use it. There's also just soooo many other packages. I like `mice` imputation; I also like functions such as `paste`", "535816": "> but it's not on Kaggle yet, so I can't use it\n\n@returnofsputnik You can install python data.table by connecting kernel to internet, I have used it on kaggle works fine. Installing from pre made wheel is recommended look this by two GM who work at h2o.ai https://www.kaggle.com/tunguz/eda-with-python-datatable#465526 and another one https://www.kaggle.com/sudalairajkumar/getting-started-with-python-datatable", "536053": "Hello bluetrain,\nThank you for your reply.\nAs I mentionned in my initial post, my english is not so good and unfortunately I can't reply to every post because of lack of time.\nThank you to everybody.\n\nOn my own, I am a beginner and I agree, kernels help to learn a lot.\nHowever, it can also discourage many kagglers who refuse by ethics maybe to fork kernels and blend it just to go up in the public leaderboard.\n \nAnyway, congratulations to you and your ABC team for your new score.\n\nIt is amazing for me ;-)\n\nGilles.", "536300": "gillou993 \nFirst of all: I think your English is good and you should not be ashamed about it ;)\n\nI must say, at the begin I had your same feeling about public kernels. I was wondering why a competition could not just be a Competition, with everyone working his best on his own (team). \nBut after a while I started appreciating kernels, as they allow me to get started in domains where I had no clue where to start from. I could grasp interesting ideas and try to improve them. I could improve my visualization skills (I am still not good at that though) etc. etc.\nI hope you will be able to get the good of public kernels too at some point.\n\nThanks for your kind words. Wish you the best for the rest of the competition!", "536628": "Hello,\nOk thank you ;-)\n\nActually, I can read english relatively easely , but it is much more complicated for me writing.\nIt is a bit frustrating because I would like to participate in discussions but it takes me too much time\nto write what I am thinking precisely...\n\nTo get back to the initial subject of this post, I agree, kernels can be a great help for beginners.\nNevertheless, public kernels may also drive beginners to the darkside of ML :-)\n\nI see a lot of kernels with hundreds or thousands of random generated features in this competition, I am a newbie but I am surprised with this gait. I am not convinced it is the right way.\n\nWish you the best for the rest of the competition too !\n\nGilles." }, "source": "meta" }