![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F274731%2F2ff5f0e7a3254fd31fe966be4df8d78d%2F2019-11-14%201.49.52.png?generation=1573711380260146&alt=media\")
I write this topic, and I know I'm gonna take a lot of downvotes.
\n\nThis is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.\nIt is really nice, and ethical to share codes, tips... in the starting phase, when everyone is not familiar with the competition.
\n\nAs the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.
\n\nWhat extremely annoys me, is that the word \"competition\" means nothing to some people, then they recklessly do whatever they like, but only in the last 15 days. Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake.
\n\nThe sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone. Then they might disappear and never share again after the competition ends.
\n\nThe thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers. They can leap to the hard-working places in a blink of a day.
\n\nEdit: \nI recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out.\nTo me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.
", "messages": [ { "id": "672455", "postDate": "11/13/2019 23:05:32", "content": "I write this topic, and I know I'm gonna take a lot of downvotes.
\n\nThis is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.\nIt is really nice, and ethical to share codes, tips... in the starting phase, when everyone is not familiar with the competition.
\n\nAs the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.
\n\nWhat extremely annoys me, is that the word \"competition\" means nothing to some people, then they recklessly do whatever they like, but only in the last 15 days. Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake.
\n\nThe sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone. Then they might disappear and never share again after the competition ends.
\n\nThe thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers. They can leap to the hard-working places in a blink of a day.
\n\nEdit: \nI recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out.\nTo me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.
", "rawMarkdown": "I write this topic, and I know I'm gonna take a lot of downvotes.\n\nThis is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.\nIt is really nice, and ethical to share codes, tips... in the starting phase, when everyone is not familiar with the competition.\n\nAs the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.\n\nWhat extremely annoys me, is that the word \"competition\" means nothing to some people, then they recklessly do whatever they like, but only in the last 15 days. Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake. \n\nThe sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone. Then they might disappear and never share again after the competition ends.\n\nThe thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers. They can leap to the hard-working places in a blink of a day.\n\nEdit: \nI recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out.\nTo me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.", "votes": null }, { "id": "672492", "postDate": "11/14/2019 00:07:23", "content": "I agree. I am out of the country right now on bereavement leave so can't compete in the last week, and can only spend a few minutes submitting ensembles of previous submissions. Seeing people openly post helper code and refinement tips in the last few days - people who should know better - is disappointing.
\n\nBut worse than any personal LB impact is that I have lost a lot of the huge respect I had for one of the most popular members of Kaggle. That's the shame of it.
", "rawMarkdown": "I agree. I am out of the country right now on bereavement leave so can't compete in the last week, and can only spend a few minutes submitting ensembles of previous submissions. Seeing people openly post helper code and refinement tips in the last few days - people who should know better - is disappointing.\n\nBut worse than any personal LB impact is that I have lost a lot of the huge respect I had for one of the most popular members of Kaggle. That's the shame of it.", "votes": null }, { "id": "672502", "postDate": "11/14/2019 00:21:22", "content": "This is a gray area. Some sharers are still aware of the competitiveness, so they just shared things which are necessary enough to help starters, such as how to effectively engineer features, how to load data faster... That's really great. But any share with high impact on LB should not be encouraged, like tips on modelling, what kind of models/features are good, or tricks to boost score directly.
", "rawMarkdown": "This is a gray area. Some sharers are still aware of the competitiveness, so they just shared things which are necessary enough to help starters, such as how to effectively engineer features, how to load data faster... That's really great. But any share with high impact on LB should not be encouraged, like tips on modelling, what kind of models/features are good, or tricks to boost score directly.", "votes": null }, { "id": "672683", "postDate": "11/14/2019 04:14:37", "content": "@robga @khahuras I hope you're not referring to me and my post \"How to Save Time and Money\" here.
\n\nMany participants including myself do not own a GPU. This makes competing very difficult. I shared how to use Google Colab a free GPU service. In order to use Google Colab efficiently you must reduce the image sizes because many times Google Colab gives you a Tesla K80 GPU and it can not handle the full images. Furthermore I provided code to help evaluate the progress of your models because once again if you're paying for GPU or struggling to work within Kaggle's 30 hour quota, you need these tricks to save time.
", "rawMarkdown": "robga @khahuras I hope you're not referring to me and my post \"How to Save Time and Money\" [here][1]. \n\nMany participants including myself do not own a GPU. This makes competing very difficult. I shared how to use Google Colab a free GPU service. In order to use Google Colab efficiently you must reduce the image sizes because many times Google Colab gives you a Tesla K80 GPU and it can not handle the full images. Furthermore I provided code to help evaluate the progress of your models because once again if you're paying for GPU or struggling to work within Kaggle's 30 hour quota, you need these tricks to save time.\n\n[1]: https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021", "votes": null }, { "id": "672689", "postDate": "11/14/2019 04:22:26", "content": "@cdeotte Of course I am not referring to you, since from past competitions, in particular, CIS, you also discouraged sharing in the last few days. Your share is truly helpful for everybody, and not devastative to top 100 teams. Ironically, the top teams are also the ones spending the most time in the competition, and they must deserve the hard work.
", "rawMarkdown": "cdeotte Of course I am not referring to you, since from past competitions, in particular, CIS, you also discouraged sharing in the last few days. Your share is truly helpful for everybody, and not devastative to top 100 teams. Ironically, the top teams are also the ones spending the most time in the competition, and they must deserve the hard work.", "votes": null }, { "id": "672711", "postDate": "11/14/2019 04:43:53", "content": "This type of discussion topic is too difficult for me...\nI recommend this discussion thread...
\n\nhttps://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846#latest-516140
", "rawMarkdown": "This type of discussion topic is too difficult for me...\nI recommend this discussion thread...\n\nhttps://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846#latest-516140", "votes": null }, { "id": "672756", "postDate": "11/14/2019 05:59:26", "content": "i suppose you are referring to me.\ni would first like to apologize if i have caused any unhappiness among kaggle community.\nit is not my intention to disrupt the LB rank, but only to share knowledge.
\n\ni do not share click and run solution. i share only ideas and partial code.
\n\ni will suggest the following system\n1. kaggle to setup a system for kaggler to file complaint of forum misuse for leaking competition secret.\n2. kaggle administration to review case by case.\n3. if kaggler are found to misuse the forum for more then e.g. N times, he will banned from competition, etc ... whatever is stated in the rule
\n\ni want to stress that information is open and fair to everyone. what may seem to be secret knowledge to one but just be common knowledge to other. i agree that this area is very gray, so with written set of rule, we will know what to do.
", "rawMarkdown": "i suppose you are referring to me.\ni would first like to apologize if i have caused any unhappiness among kaggle community.\nit is not my intention to disrupt the LB rank, but only to share knowledge.\n\ni do not share click and run solution. i share only ideas and partial code.\n\ni will suggest the following system\n1. kaggle to setup a system for kaggler to file complaint of forum misuse for leaking competition secret.\n2. kaggle administration to review case by case.\n3. if kaggler are found to misuse the forum for more then e.g. N times, he will banned from competition, etc ... whatever is stated in the rule\n\ni want to stress that information is open and fair to everyone. what may seem to be secret knowledge to one but just be common knowledge to other. i agree that this area is very gray, so with written set of rule, we will know what to do.", "votes": null }, { "id": "672757", "postDate": "11/14/2019 06:00:19", "content": "I don't think the code or models shared in the last week are different from those in the public notebooks.\nThe only thing critical here is this weird competition we can ensemble these kind of models to get into medal zones, \n(I actually found magic ensemble +0.005 weeks ago, but I didn't disclose how).
\n\n
for clarification, I mean the lb score are no difference.
", "rawMarkdown": "for clarification, I mean the lb score are no difference.", "votes": null }, { "id": "672788", "postDate": "11/14/2019 06:47:29", "content": "Please just don't share in the last 7 days. That's simply it. Anything you share, not everyone can handle. The shared things will disrupt LB, and make top teams feel really upset.\nNo apology needed. Just please understand the spirit of the competition. \nMaybe no one from Kaggle is free enough to review such cases as you propose. This is not knowledge-sharing, I emphasize again, especially in the last days.\nWhy don't you just wait until it ends and then share? Can you just go away and refrain from sharing for any competition coming to the deadline?\nDo you know this is a 2-month competition? And you came in the last 2 weeks spoiling a lot of tips, code,... To me personally, I can't take it well.
", "rawMarkdown": "Please just don't share in the last 7 days. That's simply it. Anything you share, not everyone can handle. The shared things will disrupt LB, and make top teams feel really upset.\nNo apology needed. Just please understand the spirit of the competition. \nMaybe no one from Kaggle is free enough to review such cases as you propose. This is not knowledge-sharing, I emphasize again, especially in the last days.\nWhy don't you just wait until it ends and then share? Can you just go away and refrain from sharing for any competition coming to the deadline?\nDo you know this is a 2-month competition? And you came in the last 2 weeks spoiling a lot of tips, code,... To me personally, I can't take it well.", "votes": null }, { "id": "672790", "postDate": "11/14/2019 06:49:00", "content": "I agree with you.\nWe can learn a lot of knowledge from him, but that's not after the competition.\nThat is before the competition is over.
", "rawMarkdown": "I agree with you.\nWe can learn a lot of knowledge from him, but that's not after the competition.\nThat is before the competition is over.", "votes": null }, { "id": "672803", "postDate": "11/14/2019 07:08:25", "content": "\n", "rawMarkdown": "> **Endi Niu wrote:**\n> (I actually found magic ensemble +0.005 weeks ago, but I didn't disclose how).\n> \nI imagine the scenario if any sharer discloses it, right now. Why do you keep it for yourself instead of disclosing it? Do you feel bad if someone discloses it.\nHowever, to me, I never know that there is a way to ensemble +0.005. So, you just confirmed the existence of the ensemble. People will work on that from this moment (including me, I don't know about it)\nHope there won't someone in the last day posting: \"using this extra layer in NN will help you increase CV/LB by 0.005\".", "votes": null }, { "id": "672818", "postDate": "11/14/2019 07:44:55", "content": "Endi Niu wrote:\n (I actually found magic ensemble +0.005 weeks ago, but I didn't disclose how).
\n \nI imagine the scenario if any sharer discloses it, right now. Why do you keep it for yourself instead of disclosing it? Do you feel bad if someone discloses it.\n However, to me, I never know that there is a way to ensemble +0.005. So, you just confirmed the existence of the ensemble. People will work on that from this moment (including me, I don't know about it)\n Hope there won't someone in the last day posting: \"using this extra layer in NN will help you increase CV/LB by 0.005\".
\n
I had to say, Heng's code helps a lot for me, not only professional code style, but also how debug model, how evaluate metrics. his code cannot help people click and run to get high score, you must spend hard time on it, understand it, absorb it.
\n\nfor common people like me, not cv researcher, not famous graduate, I have to study from the community. Heng's code is open for everyone, if like you said, we cannot share sometimes, can you imagine that scene, people could not share all the time
\n\nthe most important thing for kaggle is knowledge sharing, thanks to these people like Heng, Chris, Andrew and others, community have grown up.
\n\npeople love hero, and people need hero
", "rawMarkdown": "I had to say, Heng's code helps a lot for me, not only professional code style, but also how debug model, how evaluate metrics. his code cannot help people click and run to get high score, you must spend hard time on it, understand it, absorb it.\n\nfor common people like me, not cv researcher, not famous graduate, I have to study from the community. Heng's code is open for everyone, if like you said, we cannot share sometimes, can you imagine that scene, people could not share all the time\n\nthe most important thing for kaggle is knowledge sharing, thanks to these people like Heng, Chris, Andrew and others, community have grown up.\n\npeople love hero, and people need hero", "votes": null }, { "id": "672829", "postDate": "11/14/2019 07:59:54", "content": "Not in the last week. Do you understand?
", "rawMarkdown": "Not in the last week. Do you understand?", "votes": null }, { "id": "672834", "postDate": "11/14/2019 08:05:37", "content": "Thank you for your post and I am sorry for how you feel about recent discussion threads. I would like to give my perspective.
\n\n> This is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.
\n\nYes, this is a competition and we should compete hard. But this doesn't mean that we shouldn't/can't share our knowledge. Kaggle is a platform where there are people of all levels: from Begineers to Experts to Masters to GMs. Some participate to compete, some participate to learn and kaggle allows that; you can learn a lot just by going through kernels and reading discussion forums. Kaggle is not only for people in the medal zone
\n\n> As the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.
\n\nTrue, but it depends on your definition of sharing.
\n\n> Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake.
\n\nI think beginners like me are extremely grateful that we have a selfless mentor, who is willing to give constructive feedback and show how we can improve. He doesn't provide run and submit codes; he only provides skeleton codes where we are supposed to fill in the flesh. This a way better than top scoring kernels, where you can just run and submit. You are a Master now, so you may not realize the importance of having a mentor.
\n\n> The sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone.
\n\nNobody is taking anything from top teams. People share their own knowledge, not from taking others
\n\n> Then they might disappear and never share again after the competition ends.
\n\nNot true!! We have seen that in the past many many times.
\n\n> The thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers
\n\nInstead of feeling bad, you can also learn from his tips/codes and see how you can improve your model. I am pretty sure your ideas/models are different from what is shared in the forums
", "rawMarkdown": "Thank you for your post and I am sorry for how you feel about recent discussion threads. I would like to give **my perspective**.\n\n> This is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.\n\nYes, this is a competition and we should compete hard. But this doesn't mean that we shouldn't/can't share our knowledge. Kaggle is a platform where there are people of all levels: from Begineers to Experts to Masters to GMs. Some participate to compete, some participate to learn and kaggle allows that; you can learn a lot just by going through kernels and reading discussion forums. Kaggle is not only for people in the **medal zone**\n\n> As the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.\n\nTrue, but it depends on your definition of **sharing**.\n\n> Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake.\n\nI think beginners like me are extremely grateful that we have a selfless **mentor**, who is willing to **give constructive feedback** and **show** how we can improve. He doesn't provide **run and submit** codes; he only provides skeleton codes where we are supposed to fill in the flesh. This a way better than top scoring kernels, where you can just run and submit. You are a **Master** now, so you may not realize the importance of having a mentor. \n\n> The sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone. \n\nNobody is taking anything from top teams. People share their **own knowledge**, not from taking others\n\n> Then they might disappear and never share again after the competition ends.\n\nNot true!! We have seen that in the past many many times.\n\n> The thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers\n\nInstead of feeling bad, you **can also** learn from his tips/codes and see how you can improve your model. I am pretty sure your ideas/models are different from what is shared in the forums", "votes": null }, { "id": "672841", "postDate": "11/14/2019 08:16:53", "content": "In the future, you need to compete to the top 10 top 20 of a competition and then you’ll understand.
", "rawMarkdown": "In the future, you need to compete to the top 10 top 20 of a competition and then you’ll understand.", "votes": null }, { "id": "672853", "postDate": "11/14/2019 08:24:38", "content": "Yes, I look forward to that day. I understand your concern but you also have to think about beginners like me(and many others). Heng just wants to help us; his motives are nice and we should be be thankful for his efforts. Good luck for the remainder of the competition.
", "rawMarkdown": "Yes, I look forward to that day. I understand your concern but you also have to think about beginners like me(and many others). Heng just wants to help us; his motives are nice and we should be be thankful for his efforts. Good luck for the remainder of the competition.", "votes": null }, { "id": "672856", "postDate": "11/14/2019 08:29:27", "content": "You seem not to understand the problem. Please read carefully what I wrote. \nI can even help you, not Heng, when the competition finishes, not in this kind of finishing time. For reference, read my past competitions’ solutions, including full working code, discussions... for everyone.
", "rawMarkdown": "You seem not to understand the problem. Please read carefully what I wrote. \nI can even help you, not Heng, when the competition finishes, not in this kind of finishing time. For reference, read my past competitions’ solutions, including full working code, discussions... for everyone.", "votes": null }, { "id": "672870", "postDate": "11/14/2019 08:45:33", "content": "@bibek777 these are very important facts that “Kaggle is not only for people in the medal zone” and there are “selfless mentors”.
\n\nBut there is no logic that says the community benefit is any less if sharing doesn’t happen in the last week of a 12 week competition. There are other early stage competitions, and a wealth of previous competitions, to learn from.
\n\nInstead of posting a firehouse of tips on Severstal Steel, then RSNA, then Understanding Clouds each in the 11th and 12th week, why not do it in the 9th and 10th week then stop? if anything, the learning benefit to beginners would be higher as they’d have time to capitalise on the hints.
", "rawMarkdown": "bibek777 these are very important facts that “Kaggle is not only for people in the medal zone” and there are “selfless mentors”. \n\nBut there is no logic that says the community benefit is any less if sharing doesn’t happen in the last week of a 12 week competition. There are other early stage competitions, and a wealth of previous competitions, to learn from. \n\nInstead of posting a firehouse of tips on Severstal Steel, then RSNA, then Understanding Clouds each in the 11th and 12th week, why not do it in the 9th and 10th week then stop? if anything, the learning benefit to beginners would be higher as they’d have time to capitalise on the hints.", "votes": null }, { "id": "672891", "postDate": "11/14/2019 08:54:17", "content": "", "rawMarkdown": "", "votes": null }, { "id": "672912", "postDate": "11/14/2019 09:22:30", "content": "I agree that sharing in the last week or two should be avoided, I like to think not about LB places or medals but motivation. For kagglers that worked hard to get a high LB, sharing close to the end may hurt the motivation as it feels the hard work was not recognized. And as @robga mentioned some of us may not be as active in the last week making it hard to benefit from the shared information. For beginners, sharing can be a great opportunity to learn faster and increase motivation. Some balance needs to be found to benefit all.
\n\nUsually, my approach is to only share complete code in the very first weeks of competition and keep it as simple and clean as possible just to work as an experimenting framework. Then I favour more the sharing of ideas without code or just some code snippets or scripts for particular tasks like optimizing post-processing, visualization of results and so on. Finally, on the last one to two weeks sharing information that may play an important role in the final results should be avoided.
\n\nAnd of course, after the competition end is time to share and learn from top solutions.
", "rawMarkdown": "I agree that sharing in the last week or two should be avoided, I like to think not about LB places or medals but motivation. For kagglers that worked hard to get a high LB, sharing close to the end may hurt the motivation as it feels the hard work was not recognized. And as @robga mentioned some of us may not be as active in the last week making it hard to benefit from the shared information. For beginners, sharing can be a great opportunity to learn faster and increase motivation. Some balance needs to be found to benefit all. \n\nUsually, my approach is to only share complete code in the very first weeks of competition and keep it as simple and clean as possible just to work as an experimenting framework. Then I favour more the sharing of ideas without code or just some code snippets or scripts for particular tasks like optimizing post-processing, visualization of results and so on. Finally, on the last one to two weeks sharing information that may play an important role in the final results should be avoided. \n\nAnd of course, after the competition end is time to share and learn from top solutions.", "votes": null }, { "id": "672967", "postDate": "11/14/2019 10:50:10", "content": "ok here a tip: increasing your decoder capacity will increase CV/LB by 0.005 :) (sarcasm)
", "rawMarkdown": "ok here a tip: increasing your decoder capacity will increase CV/LB by 0.005 :) (sarcasm)", "votes": null }, { "id": "672990", "postDate": "11/14/2019 11:14:09", "content": "increase the fraction of CV divided by LB, right? ;)
", "rawMarkdown": "increase the fraction of CV divided by LB, right? ;)", "votes": null }, { "id": "673015", "postDate": "11/14/2019 12:01:45", "content": "@bibek777 you can always learn after the competition is over or in other competitions that are already over. No need to share information at the end of the competition
", "rawMarkdown": "bibek777 you can always learn after the competition is over or in other competitions that are already over. No need to share information at the end of the competition", "votes": null }, { "id": "673060", "postDate": "11/14/2019 13:28:15", "content": "Reminds me of https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846
", "rawMarkdown": "Reminds me of https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846", "votes": null }, { "id": "673754", "postDate": "11/15/2019 12:53:36", "content": "I concur. Here is a longer comment as the previous one was cryptic.
\n\nIt's very simple. Sharing during last week should be avoided. Why? Because it is unfair to those who cannot benefit from the sharing, for instance people away (see an example in other comments here).
\n\nWhen I participate in a competition I warn people against sharing lately, see for instance https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846, but this always lead to heated discussions.
\n\nLet me be clear: I am not against sharing, and do share a lot. Sharing is good, an great for those who want to learn. Issue is not sharing, issue is timing of sharing.
\n\nThe worst I saw was a silver medal kernel shared 6 hours before end of competition in Talking Data Fraud Detection competition. A disaster that made many kagglers stop kaggling. This is bad, and this should be avoided by all means.
", "rawMarkdown": "I concur. Here is a longer comment as the previous one was cryptic.\n\n It's very simple. Sharing during last week should be avoided. Why? Because it is unfair to those who cannot benefit from the sharing, for instance people away (see an example in other comments here). \n\nWhen I participate in a competition I warn people against sharing lately, see for instance https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846, but this always lead to heated discussions.\n\nLet me be clear: I am not against sharing, and do share a lot. Sharing is good, an great for those who want to learn. Issue is not sharing, issue is timing of sharing.\n\nThe worst I saw was a silver medal kernel shared 6 hours before end of competition in Talking Data Fraud Detection competition. A disaster that made many kagglers stop kaggling. This is bad, and this should be avoided by all means.", "votes": null }, { "id": "673844", "postDate": "11/15/2019 14:48:35", "content": "Reminds me of the Talking Data competition, my first kaggle competition where I learned a lot knowledge from you CPMP, hope to meet you in the next competition again.
", "rawMarkdown": "Reminds me of the Talking Data competition, my first kaggle competition where I learned a lot knowledge from you CPMP, hope to meet you in the next competition again.", "votes": null }, { "id": "673877", "postDate": "11/15/2019 15:34:43", "content": "Late game sharing in a deep learning competition is particularly frustrating for those that do not own GPU. It's already difficult to utilize secrets that are revealed in the last few days of a competition, but for participants that don't own a GPU it is even harder.
", "rawMarkdown": "Late game sharing in a deep learning competition is particularly frustrating for those that do not own GPU. It's already difficult to utilize secrets that are revealed in the last few days of a competition, but for participants that don't own a GPU it is even harder.", "votes": null }, { "id": "673923", "postDate": "11/15/2019 16:59:50", "content": "but the GPUs is also the part of the competition.
", "rawMarkdown": "but the GPUs is also the part of the competition.", "votes": null }, { "id": "673925", "postDate": "11/15/2019 17:01:14", "content": "agree, the issue is the timing of sharing.
", "rawMarkdown": "agree, the issue is the **timing** of sharing.", "votes": null }, { "id": "673957", "postDate": "11/15/2019 17:51:59", "content": "Many participants joined this competition before Kaggle changed from unlimited GPU to quota GPU.
", "rawMarkdown": "Many participants joined this competition before Kaggle changed from unlimited GPU to quota GPU.", "votes": null }, { "id": "673958", "postDate": "11/15/2019 17:53:30", "content": "Maybe it is time to buy a GPU ^_^
", "rawMarkdown": "Maybe it is time to buy a GPU ^_^", "votes": null }, { "id": "673959", "postDate": "11/15/2019 17:55:49", "content": "Maybe :-)
", "rawMarkdown": "Maybe :-)", "votes": null }, { "id": "674015", "postDate": "11/15/2019 19:52:09", "content": "Time to use your IEEE prize money! ^_^ @cdeotte
", "rawMarkdown": "Time to use your IEEE prize money! ^_^ @cdeotte", "votes": null }, { "id": "674263", "postDate": "11/16/2019 05:48:54", "content": "Respected Sir Kha Vo,
\n\nI am sorry but your complaint isn't valid.
\n\nAs you know, there are real human beings on this forum who are curious about the solution of the problem not about the money. The money is the byproduct. The real happiness is to get a solution close to the top 3 place.
\n\nIf you are not happy with the solutions posted near the deadline then it's better that you should not submit your solution long before deadline. You should wait until the deadline has approached. In this manner, you will be able to improve your accuracy / ranking by looking at what others are doing.
\n\nRemember, those who post solutions are just curious human beings who believe in the availability of knowledge to everyone. They are less interested in money.
\n\nAlso remember, those who have won the competition didn't do it for the money. It's for something else.
\n\nIf you have attempted to solve any puzzle in solicitude then you can understand the level of happiness that your fellow Kagglers feel, with whom you are unhappy.
\n\nI hope, my humble suggestion puts you at peace with the community.
\n\nCheers!
\n\nWaqas.
\n\nP.S.
\n\nNo I didn't down voted you!
", "rawMarkdown": "Respected Sir Kha Vo, \n\nI am sorry but your complaint isn't valid.\n\nAs you know, there are real human beings on this forum who are curious about the solution of the problem not about the money. The money is the byproduct. The real happiness is to get a solution close to the top 3 place. \n\nIf you are not happy with the solutions posted near the deadline then it's better that you should not submit your solution long before deadline. You should wait until the deadline has approached. In this manner, you will be able to improve your accuracy / ranking by looking at what others are doing. \n\nRemember, those who post solutions are just curious human beings who believe in the availability of knowledge to everyone. They are less interested in money. \n\nAlso remember, those who have won the competition didn't do it for the money. It's for something else. \n\nIf you have attempted to solve any puzzle in solicitude then you can understand the level of happiness that your fellow Kagglers feel, with whom you are unhappy. \n\nI hope, my humble suggestion puts you at peace with the community. \n\nCheers! \n\nWaqas.\n\nP.S.\n\nNo I didn't down voted you!", "votes": null }, { "id": "674280", "postDate": "11/16/2019 06:44:36", "content": "Thanks for your nice comment. But let me clarify that I don't care about money, but my rank only, which is affected by the late sharing. I spent time to find some method that boosts the score, then somebody revealed it for free, in the last few days. I noticed that you are a new Kaggler, aren't you? If so, I encourage you to stay with Kaggle and compete regularly, learn from it consistently, don't leave, and you'll understand.\nBtw, I recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out. To me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.
", "rawMarkdown": "Thanks for your nice comment. But let me clarify that I don't care about money, but my rank only, which is affected by the late sharing. I spent time to find some method that boosts the score, then somebody revealed it for free, in the last few days. I noticed that you are a new Kaggler, aren't you? If so, I encourage you to stay with Kaggle and compete regularly, learn from it consistently, don't leave, and you'll understand.\nBtw, I recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out. To me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.", "votes": null }, { "id": "674393", "postDate": "11/16/2019 12:15:14", "content": "@onlineearningwaqas who are you really? This is clearly a fake profile with no activity.
", "rawMarkdown": "onlineearningwaqas who are you really? This is clearly a fake profile with no activity.", "votes": null }, { "id": "674408", "postDate": "11/16/2019 13:00:19", "content": "Respected Sir CPMP,
\n\nI am a chemical engineer by profession.
\n\nI also work as a freelancer on Fiverr. My ID on Fiverr is the same as here.
\n\nRecently, I have been introduced to the field of data science. I'm trying to learn it.
\n\nOne of the online trainers, introduced me to this great community of data scientists.
\n\nI am still trying to learn it. I am very sorry if any of my comments have hurt you.
\n\nI consider all of the members as my seniors here. It's always my pleasure to be of any help in the field of thermodynamics, chemical process modeling, HVAC calculations, and any other form of mathematical modeling.
\n\nBest Regards,
\n\nMuhammad Waqas.
\n\nP. S.
\n\nThis is real me.
", "rawMarkdown": "Respected Sir CPMP,\n\nI am a chemical engineer by profession. \n\nI also work as a freelancer on Fiverr. My ID on Fiverr is the same as here. \n\nRecently, I have been introduced to the field of data science. I'm trying to learn it.\n\nOne of the online trainers, introduced me to this great community of data scientists. \n\nI am still trying to learn it. I am very sorry if any of my comments have hurt you. \n\nI consider all of the members as my seniors here. It's always my pleasure to be of any help in the field of thermodynamics, chemical process modeling, HVAC calculations, and any other form of mathematical modeling.\n\nBest Regards, \n\nMuhammad Waqas. \n\nP. S. \n\nThis is real me.", "votes": null }, { "id": "674451", "postDate": "11/16/2019 14:30:51", "content": "Hi, thanks for the response. I am not hurt at all, I was just surprised that you give advice to a seasoned Kaggler yet have not entered any competition. That's all, not a big deal.
\n\nGood luck on yur future competing on kaggle.
", "rawMarkdown": "Hi, thanks for the response. I am not hurt at all, I was just surprised that you give advice to a seasoned Kaggler yet have not entered any competition. That's all, not a big deal.\n\nGood luck on yur future competing on kaggle.", "votes": null }, { "id": "674535", "postDate": "11/16/2019 17:02:02", "content": "Some people take this to the next level and farm medals. This is one of the downsides of gamification and it is inevitable.
", "rawMarkdown": "Some people take this to the next level and farm medals. This is one of the downsides of gamification and it is inevitable.", "votes": null }, { "id": "676179", "postDate": "11/19/2019 02:22:23", "content": "First , I am totally in neutrality .\nSecond, actually this is a controversial topic , this problem seems appearing in so many competitions , but we don't have a court here , is there any necessary that we should set up a council consist of top 50th kaggler ranked in kaggle .
", "rawMarkdown": "First , I am totally in neutrality .\nSecond, actually this is a controversial topic , this problem seems appearing in so many competitions , but we don't have a court here , is there any necessary that we should set up a council consist of top 50th kaggler ranked in kaggle .", "votes": null }, { "id": "676184", "postDate": "11/19/2019 02:27:17", "content": "This is about balancing between passion and compassion. No one will ever be wrong by sharing with passion, but let’s do it at the right time, and consider the feelings of everybody. In the last week I did not have enough time to do all late sharing codes and ideas, as well as stamina exhausted. That’s unfair.
", "rawMarkdown": "This is about balancing between passion and compassion. No one will ever be wrong by sharing with passion, but let’s do it at the right time, and consider the feelings of everybody. In the last week I did not have enough time to do all late sharing codes and ideas, as well as stamina exhausted. That’s unfair.", "votes": null }, { "id": "676187", "postDate": "11/19/2019 02:28:30", "content": "\n\n\nwe should set up a council consist of top 50th kaggler ranked in kaggle
\n
and we shall call it Jedi Council :p
", "rawMarkdown": "> we should set up a council consist of top 50th kaggler ranked in kaggle\n\nand we shall call it Jedi Council :p", "votes": null }, { "id": "676192", "postDate": "11/19/2019 02:32:19", "content": "just don’t do unnecessary things that may hurt even 1 person. Late sharing generally is NOT a kind behaviour.
", "rawMarkdown": "just don’t do unnecessary things that may hurt even 1 person. Late sharing generally is NOT a kind behaviour.", "votes": null }, { "id": "676248", "postDate": "11/19/2019 03:35:44", "content": "Yes , I totally understand what happened on you and feel regretful about that .But I am in neutrality. What I want to say is that ,\n1 : If you regard kaggleas a competition platform , 'sharing idea in the last week' is definetly evil.\n2 :If you regard kaggle as a knowledge platform , that how to solve the problem is the most important thing but not the rank. 'sharing idea in the last week' is not evil.\nWhichever standpoint we vote for , I think the two points above are equal.
Everyone saw the public share information!
", "rawMarkdown": "Everyone saw the public share information!", "votes": null }, { "id": "860334", "postDate": "05/25/2020 07:51:02", "content": "Even if you regard kaggle as a knowledge platform, the sharing doesn't have to take place in the last week
", "rawMarkdown": "Even if you regard kaggle as a knowledge platform, the sharing doesn't have to take place in the last week", "votes": null } ], "comments": [ { "id": 672492, "author_name": "robga", "author_url": "", "post_date": "11/14/2019 00:07:23", "content": "I agree. I am out of the country right now on bereavement leave so can't compete in the last week, and can only spend a few minutes submitting ensembles of previous submissions. Seeing people openly post helper code and refinement tips in the last few days - people who should know better - is disappointing.
\n\nBut worse than any personal LB impact is that I have lost a lot of the huge respect I had for one of the most popular members of Kaggle. That's the shame of it.
", "votes": null, "replies": [ { "id": 672502, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 00:21:22", "content": "This is a gray area. Some sharers are still aware of the competitiveness, so they just shared things which are necessary enough to help starters, such as how to effectively engineer features, how to load data faster... That's really great. But any share with high impact on LB should not be encouraged, like tips on modelling, what kind of models/features are good, or tricks to boost score directly.
", "votes": null, "replies": [] } ] }, { "id": 672683, "author_name": "cdeotte", "author_url": "", "post_date": "11/14/2019 04:14:37", "content": "@robga @khahuras I hope you're not referring to me and my post \"How to Save Time and Money\" here.
\n\nMany participants including myself do not own a GPU. This makes competing very difficult. I shared how to use Google Colab a free GPU service. In order to use Google Colab efficiently you must reduce the image sizes because many times Google Colab gives you a Tesla K80 GPU and it can not handle the full images. Furthermore I provided code to help evaluate the progress of your models because once again if you're paying for GPU or struggling to work within Kaggle's 30 hour quota, you need these tricks to save time.
", "votes": null, "replies": [ { "id": 672689, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 04:22:26", "content": "@cdeotte Of course I am not referring to you, since from past competitions, in particular, CIS, you also discouraged sharing in the last few days. Your share is truly helpful for everybody, and not devastative to top 100 teams. Ironically, the top teams are also the ones spending the most time in the competition, and they must deserve the hard work.
", "votes": null, "replies": [] } ] }, { "id": 672711, "author_name": "limerobot", "author_url": "", "post_date": "11/14/2019 04:43:53", "content": "This type of discussion topic is too difficult for me...\nI recommend this discussion thread...
\n\nhttps://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846#latest-516140
", "votes": null, "replies": [] }, { "id": 672756, "author_name": "hengck23", "author_url": "", "post_date": "11/14/2019 05:59:26", "content": "i suppose you are referring to me.\ni would first like to apologize if i have caused any unhappiness among kaggle community.\nit is not my intention to disrupt the LB rank, but only to share knowledge.
\n\ni do not share click and run solution. i share only ideas and partial code.
\n\ni will suggest the following system\n1. kaggle to setup a system for kaggler to file complaint of forum misuse for leaking competition secret.\n2. kaggle administration to review case by case.\n3. if kaggler are found to misuse the forum for more then e.g. N times, he will banned from competition, etc ... whatever is stated in the rule
\n\ni want to stress that information is open and fair to everyone. what may seem to be secret knowledge to one but just be common knowledge to other. i agree that this area is very gray, so with written set of rule, we will know what to do.
", "votes": null, "replies": [ { "id": 672788, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 06:47:29", "content": "Please just don't share in the last 7 days. That's simply it. Anything you share, not everyone can handle. The shared things will disrupt LB, and make top teams feel really upset.\nNo apology needed. Just please understand the spirit of the competition. \nMaybe no one from Kaggle is free enough to review such cases as you propose. This is not knowledge-sharing, I emphasize again, especially in the last days.\nWhy don't you just wait until it ends and then share? Can you just go away and refrain from sharing for any competition coming to the deadline?\nDo you know this is a 2-month competition? And you came in the last 2 weeks spoiling a lot of tips, code,... To me personally, I can't take it well.
", "votes": null, "replies": [] }, { "id": 672818, "author_name": "jt120lz", "author_url": "", "post_date": "11/14/2019 07:44:55", "content": "I had to say, Heng's code helps a lot for me, not only professional code style, but also how debug model, how evaluate metrics. his code cannot help people click and run to get high score, you must spend hard time on it, understand it, absorb it.
\n\nfor common people like me, not cv researcher, not famous graduate, I have to study from the community. Heng's code is open for everyone, if like you said, we cannot share sometimes, can you imagine that scene, people could not share all the time
\n\nthe most important thing for kaggle is knowledge sharing, thanks to these people like Heng, Chris, Andrew and others, community have grown up.
\n\npeople love hero, and people need hero
", "votes": null, "replies": [] }, { "id": 672829, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 07:59:54", "content": "Not in the last week. Do you understand?
", "votes": null, "replies": [] }, { "id": 685613, "author_name": "cswwp347724", "author_url": "", "post_date": "12/02/2019 01:59:32", "content": "Everyone saw the public share information!
", "votes": null, "replies": [] } ] }, { "id": 672757, "author_name": "niuddd", "author_url": "", "post_date": "11/14/2019 06:00:19", "content": "I don't think the code or models shared in the last week are different from those in the public notebooks.\nThe only thing critical here is this weird competition we can ensemble these kind of models to get into medal zones, \n(I actually found magic ensemble +0.005 weeks ago, but I didn't disclose how).
\n\n
for clarification, I mean the lb score are no difference.
", "votes": null, "replies": [] }, { "id": 672803, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 07:08:25", "content": "\n", "votes": null, "replies": [] }, { "id": 672967, "author_name": "tugstugi", "author_url": "", "post_date": "11/14/2019 10:50:10", "content": "Endi Niu wrote:\n (I actually found magic ensemble +0.005 weeks ago, but I didn't disclose how).
\n \nI imagine the scenario if any sharer discloses it, right now. Why do you keep it for yourself instead of disclosing it? Do you feel bad if someone discloses it.\n However, to me, I never know that there is a way to ensemble +0.005. So, you just confirmed the existence of the ensemble. People will work on that from this moment (including me, I don't know about it)\n Hope there won't someone in the last day posting: \"using this extra layer in NN will help you increase CV/LB by 0.005\".
\n
ok here a tip: increasing your decoder capacity will increase CV/LB by 0.005 :) (sarcasm)
", "votes": null, "replies": [] }, { "id": 672990, "author_name": "niuddd", "author_url": "", "post_date": "11/14/2019 11:14:09", "content": "increase the fraction of CV divided by LB, right? ;)
", "votes": null, "replies": [] } ] }, { "id": 672790, "author_name": "gwanghan", "author_url": "", "post_date": "11/14/2019 06:49:00", "content": "I agree with you.\nWe can learn a lot of knowledge from him, but that's not after the competition.\nThat is before the competition is over.
", "votes": null, "replies": [] }, { "id": 672834, "author_name": "bibek777", "author_url": "", "post_date": "11/14/2019 08:05:37", "content": "Thank you for your post and I am sorry for how you feel about recent discussion threads. I would like to give my perspective.
\n\n> This is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.
\n\nYes, this is a competition and we should compete hard. But this doesn't mean that we shouldn't/can't share our knowledge. Kaggle is a platform where there are people of all levels: from Begineers to Experts to Masters to GMs. Some participate to compete, some participate to learn and kaggle allows that; you can learn a lot just by going through kernels and reading discussion forums. Kaggle is not only for people in the medal zone
\n\n> As the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.
\n\nTrue, but it depends on your definition of sharing.
\n\n> Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake.
\n\nI think beginners like me are extremely grateful that we have a selfless mentor, who is willing to give constructive feedback and show how we can improve. He doesn't provide run and submit codes; he only provides skeleton codes where we are supposed to fill in the flesh. This a way better than top scoring kernels, where you can just run and submit. You are a Master now, so you may not realize the importance of having a mentor.
\n\n> The sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone.
\n\nNobody is taking anything from top teams. People share their own knowledge, not from taking others
\n\n> Then they might disappear and never share again after the competition ends.
\n\nNot true!! We have seen that in the past many many times.
\n\n> The thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers
\n\nInstead of feeling bad, you can also learn from his tips/codes and see how you can improve your model. I am pretty sure your ideas/models are different from what is shared in the forums
", "votes": null, "replies": [ { "id": 672841, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 08:16:53", "content": "In the future, you need to compete to the top 10 top 20 of a competition and then you’ll understand.
", "votes": null, "replies": [] }, { "id": 672853, "author_name": "bibek777", "author_url": "", "post_date": "11/14/2019 08:24:38", "content": "Yes, I look forward to that day. I understand your concern but you also have to think about beginners like me(and many others). Heng just wants to help us; his motives are nice and we should be be thankful for his efforts. Good luck for the remainder of the competition.
", "votes": null, "replies": [] }, { "id": 672856, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 08:29:27", "content": "You seem not to understand the problem. Please read carefully what I wrote. \nI can even help you, not Heng, when the competition finishes, not in this kind of finishing time. For reference, read my past competitions’ solutions, including full working code, discussions... for everyone.
", "votes": null, "replies": [] }, { "id": 672870, "author_name": "robga", "author_url": "", "post_date": "11/14/2019 08:45:33", "content": "@bibek777 these are very important facts that “Kaggle is not only for people in the medal zone” and there are “selfless mentors”.
\n\nBut there is no logic that says the community benefit is any less if sharing doesn’t happen in the last week of a 12 week competition. There are other early stage competitions, and a wealth of previous competitions, to learn from.
\n\nInstead of posting a firehouse of tips on Severstal Steel, then RSNA, then Understanding Clouds each in the 11th and 12th week, why not do it in the 9th and 10th week then stop? if anything, the learning benefit to beginners would be higher as they’d have time to capitalise on the hints.
", "votes": null, "replies": [] }, { "id": 672891, "author_name": "khahuras", "author_url": "", "post_date": "11/14/2019 08:54:17", "content": "", "votes": null, "replies": [] }, { "id": 673015, "author_name": "igormunizims", "author_url": "", "post_date": "11/14/2019 12:01:45", "content": "@bibek777 you can always learn after the competition is over or in other competitions that are already over. No need to share information at the end of the competition
", "votes": null, "replies": [] } ] }, { "id": 672912, "author_name": "mnpinto", "author_url": "", "post_date": "11/14/2019 09:22:30", "content": "I agree that sharing in the last week or two should be avoided, I like to think not about LB places or medals but motivation. For kagglers that worked hard to get a high LB, sharing close to the end may hurt the motivation as it feels the hard work was not recognized. And as @robga mentioned some of us may not be as active in the last week making it hard to benefit from the shared information. For beginners, sharing can be a great opportunity to learn faster and increase motivation. Some balance needs to be found to benefit all.
\n\nUsually, my approach is to only share complete code in the very first weeks of competition and keep it as simple and clean as possible just to work as an experimenting framework. Then I favour more the sharing of ideas without code or just some code snippets or scripts for particular tasks like optimizing post-processing, visualization of results and so on. Finally, on the last one to two weeks sharing information that may play an important role in the final results should be avoided.
\n\nAnd of course, after the competition end is time to share and learn from top solutions.
", "votes": null, "replies": [] }, { "id": 673060, "author_name": "cpmpml", "author_url": "", "post_date": "11/14/2019 13:28:15", "content": "Reminds me of https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846
", "votes": null, "replies": [] }, { "id": 673754, "author_name": "cpmpml", "author_url": "", "post_date": "11/15/2019 12:53:36", "content": "I concur. Here is a longer comment as the previous one was cryptic.
\n\nIt's very simple. Sharing during last week should be avoided. Why? Because it is unfair to those who cannot benefit from the sharing, for instance people away (see an example in other comments here).
\n\nWhen I participate in a competition I warn people against sharing lately, see for instance https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846, but this always lead to heated discussions.
\n\nLet me be clear: I am not against sharing, and do share a lot. Sharing is good, an great for those who want to learn. Issue is not sharing, issue is timing of sharing.
\n\nThe worst I saw was a silver medal kernel shared 6 hours before end of competition in Talking Data Fraud Detection competition. A disaster that made many kagglers stop kaggling. This is bad, and this should be avoided by all means.
", "votes": null, "replies": [ { "id": 673844, "author_name": "niuddd", "author_url": "", "post_date": "11/15/2019 14:48:35", "content": "Reminds me of the Talking Data competition, my first kaggle competition where I learned a lot knowledge from you CPMP, hope to meet you in the next competition again.
", "votes": null, "replies": [] }, { "id": 673925, "author_name": "garybios", "author_url": "", "post_date": "11/15/2019 17:01:14", "content": "agree, the issue is the timing of sharing.
", "votes": null, "replies": [] } ] }, { "id": 673877, "author_name": "cdeotte", "author_url": "", "post_date": "11/15/2019 15:34:43", "content": "Late game sharing in a deep learning competition is particularly frustrating for those that do not own GPU. It's already difficult to utilize secrets that are revealed in the last few days of a competition, but for participants that don't own a GPU it is even harder.
", "votes": null, "replies": [ { "id": 673923, "author_name": "garybios", "author_url": "", "post_date": "11/15/2019 16:59:50", "content": "but the GPUs is also the part of the competition.
", "votes": null, "replies": [] }, { "id": 673957, "author_name": "cdeotte", "author_url": "", "post_date": "11/15/2019 17:51:59", "content": "Many participants joined this competition before Kaggle changed from unlimited GPU to quota GPU.
", "votes": null, "replies": [] }, { "id": 673958, "author_name": "naivelamb", "author_url": "", "post_date": "11/15/2019 17:53:30", "content": "Maybe it is time to buy a GPU ^_^
", "votes": null, "replies": [] }, { "id": 673959, "author_name": "cdeotte", "author_url": "", "post_date": "11/15/2019 17:55:49", "content": "Maybe :-)
", "votes": null, "replies": [] }, { "id": 674015, "author_name": "naivelamb", "author_url": "", "post_date": "11/15/2019 19:52:09", "content": "Time to use your IEEE prize money! ^_^ @cdeotte
", "votes": null, "replies": [] } ] }, { "id": 674263, "author_name": "onlineearningwaqas", "author_url": "", "post_date": "11/16/2019 05:48:54", "content": "Respected Sir Kha Vo,
\n\nI am sorry but your complaint isn't valid.
\n\nAs you know, there are real human beings on this forum who are curious about the solution of the problem not about the money. The money is the byproduct. The real happiness is to get a solution close to the top 3 place.
\n\nIf you are not happy with the solutions posted near the deadline then it's better that you should not submit your solution long before deadline. You should wait until the deadline has approached. In this manner, you will be able to improve your accuracy / ranking by looking at what others are doing.
\n\nRemember, those who post solutions are just curious human beings who believe in the availability of knowledge to everyone. They are less interested in money.
\n\nAlso remember, those who have won the competition didn't do it for the money. It's for something else.
\n\nIf you have attempted to solve any puzzle in solicitude then you can understand the level of happiness that your fellow Kagglers feel, with whom you are unhappy.
\n\nI hope, my humble suggestion puts you at peace with the community.
\n\nCheers!
\n\nWaqas.
\n\nP.S.
\n\nNo I didn't down voted you!
", "votes": null, "replies": [ { "id": 674280, "author_name": "khahuras", "author_url": "", "post_date": "11/16/2019 06:44:36", "content": "Thanks for your nice comment. But let me clarify that I don't care about money, but my rank only, which is affected by the late sharing. I spent time to find some method that boosts the score, then somebody revealed it for free, in the last few days. I noticed that you are a new Kaggler, aren't you? If so, I encourage you to stay with Kaggle and compete regularly, learn from it consistently, don't leave, and you'll understand.\nBtw, I recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out. To me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.
", "votes": null, "replies": [] }, { "id": 674393, "author_name": "cpmpml", "author_url": "", "post_date": "11/16/2019 12:15:14", "content": "@onlineearningwaqas who are you really? This is clearly a fake profile with no activity.
", "votes": null, "replies": [] }, { "id": 674408, "author_name": "onlineearningwaqas", "author_url": "", "post_date": "11/16/2019 13:00:19", "content": "Respected Sir CPMP,
\n\nI am a chemical engineer by profession.
\n\nI also work as a freelancer on Fiverr. My ID on Fiverr is the same as here.
\n\nRecently, I have been introduced to the field of data science. I'm trying to learn it.
\n\nOne of the online trainers, introduced me to this great community of data scientists.
\n\nI am still trying to learn it. I am very sorry if any of my comments have hurt you.
\n\nI consider all of the members as my seniors here. It's always my pleasure to be of any help in the field of thermodynamics, chemical process modeling, HVAC calculations, and any other form of mathematical modeling.
\n\nBest Regards,
\n\nMuhammad Waqas.
\n\nP. S.
\n\nThis is real me.
", "votes": null, "replies": [] }, { "id": 674451, "author_name": "cpmpml", "author_url": "", "post_date": "11/16/2019 14:30:51", "content": "Hi, thanks for the response. I am not hurt at all, I was just surprised that you give advice to a seasoned Kaggler yet have not entered any competition. That's all, not a big deal.
\n\nGood luck on yur future competing on kaggle.
", "votes": null, "replies": [] } ] }, { "id": 674535, "author_name": "gunesevitan", "author_url": "", "post_date": "11/16/2019 17:02:02", "content": "Some people take this to the next level and farm medals. This is one of the downsides of gamification and it is inevitable.
", "votes": null, "replies": [] }, { "id": 676179, "author_name": "xujingzhao", "author_url": "", "post_date": "11/19/2019 02:22:23", "content": "First , I am totally in neutrality .\nSecond, actually this is a controversial topic , this problem seems appearing in so many competitions , but we don't have a court here , is there any necessary that we should set up a council consist of top 50th kaggler ranked in kaggle .
", "votes": null, "replies": [ { "id": 676184, "author_name": "khahuras", "author_url": "", "post_date": "11/19/2019 02:27:17", "content": "This is about balancing between passion and compassion. No one will ever be wrong by sharing with passion, but let’s do it at the right time, and consider the feelings of everybody. In the last week I did not have enough time to do all late sharing codes and ideas, as well as stamina exhausted. That’s unfair.
", "votes": null, "replies": [] }, { "id": 676187, "author_name": "nyleve", "author_url": "", "post_date": "11/19/2019 02:28:30", "content": "\n\n\nwe should set up a council consist of top 50th kaggler ranked in kaggle
\n
and we shall call it Jedi Council :p
", "votes": null, "replies": [] }, { "id": 676192, "author_name": "khahuras", "author_url": "", "post_date": "11/19/2019 02:32:19", "content": "just don’t do unnecessary things that may hurt even 1 person. Late sharing generally is NOT a kind behaviour.
", "votes": null, "replies": [] }, { "id": 676248, "author_name": "xujingzhao", "author_url": "", "post_date": "11/19/2019 03:35:44", "content": "Yes , I totally understand what happened on you and feel regretful about that .But I am in neutrality. What I want to say is that ,\n1 : If you regard kaggleas a competition platform , 'sharing idea in the last week' is definetly evil.\n2 :If you regard kaggle as a knowledge platform , that how to solve the problem is the most important thing but not the rank. 'sharing idea in the last week' is not evil.\nWhichever standpoint we vote for , I think the two points above are equal.
Even if you regard kaggle as a knowledge platform, the sharing doesn't have to take place in the last week
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "672455": "I write this topic, and I know I'm gonna take a lot of downvotes.\n\nThis is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.\nIt is really nice, and ethical to share codes, tips... in the starting phase, when everyone is not familiar with the competition.\n\nAs the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.\n\nWhat extremely annoys me, is that the word \"competition\" means nothing to some people, then they recklessly do whatever they like, but only in the last 15 days. Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake. \n\nThe sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone. Then they might disappear and never share again after the competition ends.\n\nThe thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers. They can leap to the hard-working places in a blink of a day.\n\nEdit: \nI recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out.\nTo me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.", "672492": "I agree. I am out of the country right now on bereavement leave so can't compete in the last week, and can only spend a few minutes submitting ensembles of previous submissions. Seeing people openly post helper code and refinement tips in the last few days - people who should know better - is disappointing.\n\nBut worse than any personal LB impact is that I have lost a lot of the huge respect I had for one of the most popular members of Kaggle. That's the shame of it.", "672502": "This is a gray area. Some sharers are still aware of the competitiveness, so they just shared things which are necessary enough to help starters, such as how to effectively engineer features, how to load data faster... That's really great. But any share with high impact on LB should not be encouraged, like tips on modelling, what kind of models/features are good, or tricks to boost score directly.", "672683": "robga @khahuras I hope you're not referring to me and my post \"How to Save Time and Money\" [here][1]. \n\nMany participants including myself do not own a GPU. This makes competing very difficult. I shared how to use Google Colab a free GPU service. In order to use Google Colab efficiently you must reduce the image sizes because many times Google Colab gives you a Tesla K80 GPU and it can not handle the full images. Furthermore I provided code to help evaluate the progress of your models because once again if you're paying for GPU or struggling to work within Kaggle's 30 hour quota, you need these tricks to save time.\n\n[1]: https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021", "672689": "cdeotte Of course I am not referring to you, since from past competitions, in particular, CIS, you also discouraged sharing in the last few days. Your share is truly helpful for everybody, and not devastative to top 100 teams. Ironically, the top teams are also the ones spending the most time in the competition, and they must deserve the hard work.", "672711": "This type of discussion topic is too difficult for me...\nI recommend this discussion thread...\n\nhttps://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846#latest-516140", "672756": "i suppose you are referring to me.\ni would first like to apologize if i have caused any unhappiness among kaggle community.\nit is not my intention to disrupt the LB rank, but only to share knowledge.\n\ni do not share click and run solution. i share only ideas and partial code.\n\ni will suggest the following system\n1. kaggle to setup a system for kaggler to file complaint of forum misuse for leaking competition secret.\n2. kaggle administration to review case by case.\n3. if kaggler are found to misuse the forum for more then e.g. N times, he will banned from competition, etc ... whatever is stated in the rule\n\ni want to stress that information is open and fair to everyone. what may seem to be secret knowledge to one but just be common knowledge to other. i agree that this area is very gray, so with written set of rule, we will know what to do.", "672757": "I don't think the code or models shared in the last week are different from those in the public notebooks.\nThe only thing critical here is this weird competition we can ensemble these kind of models to get into medal zones, \n(I actually found magic ensemble +0.005 weeks ago, but I didn't disclose how).\n\n", "672760": "for clarification, I mean the lb score are no difference.", "672788": "Please just don't share in the last 7 days. That's simply it. Anything you share, not everyone can handle. The shared things will disrupt LB, and make top teams feel really upset.\nNo apology needed. Just please understand the spirit of the competition. \nMaybe no one from Kaggle is free enough to review such cases as you propose. This is not knowledge-sharing, I emphasize again, especially in the last days.\nWhy don't you just wait until it ends and then share? Can you just go away and refrain from sharing for any competition coming to the deadline?\nDo you know this is a 2-month competition? And you came in the last 2 weeks spoiling a lot of tips, code,... To me personally, I can't take it well.", "672790": "I agree with you.\nWe can learn a lot of knowledge from him, but that's not after the competition.\nThat is before the competition is over.", "672803": "> **Endi Niu wrote:**\n> (I actually found magic ensemble +0.005 weeks ago, but I didn't disclose how).\n> \nI imagine the scenario if any sharer discloses it, right now. Why do you keep it for yourself instead of disclosing it? Do you feel bad if someone discloses it.\nHowever, to me, I never know that there is a way to ensemble +0.005. So, you just confirmed the existence of the ensemble. People will work on that from this moment (including me, I don't know about it)\nHope there won't someone in the last day posting: \"using this extra layer in NN will help you increase CV/LB by 0.005\".", "672818": "I had to say, Heng's code helps a lot for me, not only professional code style, but also how debug model, how evaluate metrics. his code cannot help people click and run to get high score, you must spend hard time on it, understand it, absorb it.\n\nfor common people like me, not cv researcher, not famous graduate, I have to study from the community. Heng's code is open for everyone, if like you said, we cannot share sometimes, can you imagine that scene, people could not share all the time\n\nthe most important thing for kaggle is knowledge sharing, thanks to these people like Heng, Chris, Andrew and others, community have grown up.\n\npeople love hero, and people need hero", "672829": "Not in the last week. Do you understand?", "672834": "Thank you for your post and I am sorry for how you feel about recent discussion threads. I would like to give **my perspective**.\n\n> This is not a knowledge sharing business, but a competition, and everyone is working hard to get a good result.\n\nYes, this is a competition and we should compete hard. But this doesn't mean that we shouldn't/can't share our knowledge. Kaggle is a platform where there are people of all levels: from Begineers to Experts to Masters to GMs. Some participate to compete, some participate to learn and kaggle allows that; you can learn a lot just by going through kernels and reading discussion forums. Kaggle is not only for people in the **medal zone**\n\n> As the competition goes to an end, sharing should be stopped. That's for the fairness, especially in this only final days.\n\nTrue, but it depends on your definition of **sharing**.\n\n> Everyone asks sharers, sharers happy to reply, including offering a lot of tips, even code, that other top teams might be keeping as secrets for their own sake.\n\nI think beginners like me are extremely grateful that we have a selfless **mentor**, who is willing to **give constructive feedback** and **show** how we can improve. He doesn't provide **run and submit** codes; he only provides skeleton codes where we are supposed to fill in the flesh. This a way better than top scoring kernels, where you can just run and submit. You are a **Master** now, so you may not realize the importance of having a mentor. \n\n> The sharers are like the \"competition balancers\", who can balance between high-placed teams and lower-placed teams by passing top tips from the rich to the poor, and make the society fairer to everyone. \n\nNobody is taking anything from top teams. People share their **own knowledge**, not from taking others\n\n> Then they might disappear and never share again after the competition ends.\n\nNot true!! We have seen that in the past many many times.\n\n> The thing is, people working hard, like me, really feel upset and treated badly, while people with no advantages may feel resurrected, with the existence of the sharers\n\nInstead of feeling bad, you **can also** learn from his tips/codes and see how you can improve your model. I am pretty sure your ideas/models are different from what is shared in the forums", "672841": "In the future, you need to compete to the top 10 top 20 of a competition and then you’ll understand.", "672853": "Yes, I look forward to that day. I understand your concern but you also have to think about beginners like me(and many others). Heng just wants to help us; his motives are nice and we should be be thankful for his efforts. Good luck for the remainder of the competition.", "672856": "You seem not to understand the problem. Please read carefully what I wrote. \nI can even help you, not Heng, when the competition finishes, not in this kind of finishing time. For reference, read my past competitions’ solutions, including full working code, discussions... for everyone.", "672870": "bibek777 these are very important facts that “Kaggle is not only for people in the medal zone” and there are “selfless mentors”. \n\nBut there is no logic that says the community benefit is any less if sharing doesn’t happen in the last week of a 12 week competition. There are other early stage competitions, and a wealth of previous competitions, to learn from. \n\nInstead of posting a firehouse of tips on Severstal Steel, then RSNA, then Understanding Clouds each in the 11th and 12th week, why not do it in the 9th and 10th week then stop? if anything, the learning benefit to beginners would be higher as they’d have time to capitalise on the hints.", "672891": "", "672912": "I agree that sharing in the last week or two should be avoided, I like to think not about LB places or medals but motivation. For kagglers that worked hard to get a high LB, sharing close to the end may hurt the motivation as it feels the hard work was not recognized. And as @robga mentioned some of us may not be as active in the last week making it hard to benefit from the shared information. For beginners, sharing can be a great opportunity to learn faster and increase motivation. Some balance needs to be found to benefit all. \n\nUsually, my approach is to only share complete code in the very first weeks of competition and keep it as simple and clean as possible just to work as an experimenting framework. Then I favour more the sharing of ideas without code or just some code snippets or scripts for particular tasks like optimizing post-processing, visualization of results and so on. Finally, on the last one to two weeks sharing information that may play an important role in the final results should be avoided. \n\nAnd of course, after the competition end is time to share and learn from top solutions.", "672967": "ok here a tip: increasing your decoder capacity will increase CV/LB by 0.005 :) (sarcasm)", "672990": "increase the fraction of CV divided by LB, right? ;)", "673015": "bibek777 you can always learn after the competition is over or in other competitions that are already over. No need to share information at the end of the competition", "673060": "Reminds me of https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846", "673754": "I concur. Here is a longer comment as the previous one was cryptic.\n\n It's very simple. Sharing during last week should be avoided. Why? Because it is unfair to those who cannot benefit from the sharing, for instance people away (see an example in other comments here). \n\nWhen I participate in a competition I warn people against sharing lately, see for instance https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/87846, but this always lead to heated discussions.\n\nLet me be clear: I am not against sharing, and do share a lot. Sharing is good, an great for those who want to learn. Issue is not sharing, issue is timing of sharing.\n\nThe worst I saw was a silver medal kernel shared 6 hours before end of competition in Talking Data Fraud Detection competition. A disaster that made many kagglers stop kaggling. This is bad, and this should be avoided by all means.", "673844": "Reminds me of the Talking Data competition, my first kaggle competition where I learned a lot knowledge from you CPMP, hope to meet you in the next competition again.", "673877": "Late game sharing in a deep learning competition is particularly frustrating for those that do not own GPU. It's already difficult to utilize secrets that are revealed in the last few days of a competition, but for participants that don't own a GPU it is even harder.", "673923": "but the GPUs is also the part of the competition.", "673925": "agree, the issue is the **timing** of sharing.", "673957": "Many participants joined this competition before Kaggle changed from unlimited GPU to quota GPU.", "673958": "Maybe it is time to buy a GPU ^_^", "673959": "Maybe :-)", "674015": "Time to use your IEEE prize money! ^_^ @cdeotte", "674263": "Respected Sir Kha Vo, \n\nI am sorry but your complaint isn't valid.\n\nAs you know, there are real human beings on this forum who are curious about the solution of the problem not about the money. The money is the byproduct. The real happiness is to get a solution close to the top 3 place. \n\nIf you are not happy with the solutions posted near the deadline then it's better that you should not submit your solution long before deadline. You should wait until the deadline has approached. In this manner, you will be able to improve your accuracy / ranking by looking at what others are doing. \n\nRemember, those who post solutions are just curious human beings who believe in the availability of knowledge to everyone. They are less interested in money. \n\nAlso remember, those who have won the competition didn't do it for the money. It's for something else. \n\nIf you have attempted to solve any puzzle in solicitude then you can understand the level of happiness that your fellow Kagglers feel, with whom you are unhappy. \n\nI hope, my humble suggestion puts you at peace with the community. \n\nCheers! \n\nWaqas.\n\nP.S.\n\nNo I didn't down voted you!", "674280": "Thanks for your nice comment. But let me clarify that I don't care about money, but my rank only, which is affected by the late sharing. I spent time to find some method that boosts the score, then somebody revealed it for free, in the last few days. I noticed that you are a new Kaggler, aren't you? If so, I encourage you to stay with Kaggle and compete regularly, learn from it consistently, don't leave, and you'll understand.\nBtw, I recently received an email from a Kaggle competition Master, in which he thanks me for posting this topic. He stated that in mostly computer vision competitions, the excessive sharing, especially in final days, heavily discouraged him from competing, but he has not the bravery to say it out. To me, for tabular competitions with tree-based methods as main modelling tools, it is even more serious for late-sharing, since LightGBM works in split seconds.", "674393": "onlineearningwaqas who are you really? This is clearly a fake profile with no activity.", "674408": "Respected Sir CPMP,\n\nI am a chemical engineer by profession. \n\nI also work as a freelancer on Fiverr. My ID on Fiverr is the same as here. \n\nRecently, I have been introduced to the field of data science. I'm trying to learn it.\n\nOne of the online trainers, introduced me to this great community of data scientists. \n\nI am still trying to learn it. I am very sorry if any of my comments have hurt you. \n\nI consider all of the members as my seniors here. It's always my pleasure to be of any help in the field of thermodynamics, chemical process modeling, HVAC calculations, and any other form of mathematical modeling.\n\nBest Regards, \n\nMuhammad Waqas. \n\nP. S. \n\nThis is real me.", "674451": "Hi, thanks for the response. I am not hurt at all, I was just surprised that you give advice to a seasoned Kaggler yet have not entered any competition. That's all, not a big deal.\n\nGood luck on yur future competing on kaggle.", "674535": "Some people take this to the next level and farm medals. This is one of the downsides of gamification and it is inevitable.", "676179": "First , I am totally in neutrality .\nSecond, actually this is a controversial topic , this problem seems appearing in so many competitions , but we don't have a court here , is there any necessary that we should set up a council consist of top 50th kaggler ranked in kaggle .", "676184": "This is about balancing between passion and compassion. No one will ever be wrong by sharing with passion, but let’s do it at the right time, and consider the feelings of everybody. In the last week I did not have enough time to do all late sharing codes and ideas, as well as stamina exhausted. That’s unfair.", "676187": "> we should set up a council consist of top 50th kaggler ranked in kaggle\n\nand we shall call it Jedi Council :p", "676192": "just don’t do unnecessary things that may hurt even 1 person. Late sharing generally is NOT a kind behaviour.", "676248": "Yes , I totally understand what happened on you and feel regretful about that .But I am in neutrality. What I want to say is that ,\n1 : If you regard `kaggle `as a competition platform , 'sharing idea in the last week' is definetly evil.\n2 :If you regard `kaggle` as a knowledge platform , that how to solve the problem is the most important thing but not the rank. 'sharing idea in the last week' is not evil.\nWhichever standpoint we vote for , I think the two points above are equal.", "685613": "Everyone saw the public share information!", "860334": "Even if you regard kaggle as a knowledge platform, the sharing doesn't have to take place in the last week" }, "source": "meta" }