Dear fellow Kaggle Competitor,
\n\nWant to have your solution count and at the same time help save lives by contributing to effective altruism??
\n\nIf so, we’d like to invite you to join a cooperative effort using ensemble methods for the Diabetic Retinopathy competition ending on Monday and the Avito Context Ad Clicks context ending on Tuesday and help win a prize that would be donated to Charity. Ensemble methods have proven time and again to win machine learning competitions and by pooling as many teams as possible, we greatly increase our chances of winning. If we do win, it will be announced at the Effective Altruism Global summit next week, and proper credit will be given to everyone who contributes.
\n\nKaggle’s rules on sharing code and data privately are strict. However they do allow for all such data to be shared publicly and made available to all contestants. With this in mind, if you are interested please pay attention to the following:
\n\nWe encourage you to also read up on Effective Altruism, “a philosophy and social movement that applies evidence and reason to determine the most effective ways to improve the world”. Any prize money won will be donated to charities recommended by The Life You Can Save and Give Well. We are happy to hear about any other requests or suggestions.
\n\nAs a last comment, by pooling our solutions together we will also be able to provide a much higher quality solution to the California Healthcare Foundation instead of them just having access to the top few. We also look forward to future cooperative efforts for Kaggle and we will be updating this blog periodically.
\n\nPlease contact us with any questions.
", "messages": [ { "id": "87132", "postDate": "07/27/2015 05:23:32", "content": "Dear fellow Kaggle Competitor,
\n\nWant to have your solution count and at the same time help save lives by contributing to effective altruism??
\n\nIf so, we’d like to invite you to join a cooperative effort using ensemble methods for the Diabetic Retinopathy competition ending on Monday and the Avito Context Ad Clicks context ending on Tuesday and help win a prize that would be donated to Charity. Ensemble methods have proven time and again to win machine learning competitions and by pooling as many teams as possible, we greatly increase our chances of winning. If we do win, it will be announced at the Effective Altruism Global summit next week, and proper credit will be given to everyone who contributes.
\n\nKaggle’s rules on sharing code and data privately are strict. However they do allow for all such data to be shared publicly and made available to all contestants. With this in mind, if you are interested please pay attention to the following:
\n\nWe encourage you to also read up on Effective Altruism, “a philosophy and social movement that applies evidence and reason to determine the most effective ways to improve the world”. Any prize money won will be donated to charities recommended by The Life You Can Save and Give Well. We are happy to hear about any other requests or suggestions.
\n\nAs a last comment, by pooling our solutions together we will also be able to provide a much higher quality solution to the California Healthcare Foundation instead of them just having access to the top few. We also look forward to future cooperative efforts for Kaggle and we will be updating this blog periodically.
\n\nPlease contact us with any questions.
", "rawMarkdown": "Dear fellow Kaggle Competitor,\r\n\r\nWant to have your solution count and at the same time help save lives by contributing to effective altruism??\r\n\r\nIf so, we’d like to invite you to join a cooperative effort using ensemble methods for the Diabetic Retinopathy competition ending on Monday and the Avito Context Ad Clicks context ending on Tuesday and help win a prize that would be donated to Charity. Ensemble methods have proven time and again to win machine learning competitions and by pooling as many teams as possible, we greatly increase our chances of winning. If we do win, it will be announced at the [Effective Altruism Global][1] summit next week, and proper credit will be given to everyone who contributes.\r\n\r\nKaggle’s rules on sharing code and data privately are strict. However they do allow for all such data to be shared publicly and made available to all contestants. With this in mind, if you are interested please pay attention to the following:\r\n\r\n 1. As soon as you are able, share the outputs of both the training and testing sets on your github (or I can provide somewhere to share this) with a LICENSE file saying “Permission is granted to use this in a competition only if the winnings are donated entirely to charity”.\r\n 2. If your results are useful then, before Monday’s deadline, you will open source your code with an MIT license (without any restriction).\r\n\r\nWe encourage you to also read up on [Effective Altruism][2], “a philosophy and social movement that applies evidence and reason to determine the most effective ways to improve the world”. Any prize money won will be donated to charities recommended by [The Life You Can Save][3] and [Give Well][4]. We are happy to hear about any other requests or suggestions.\r\n\r\nAs a last comment, by pooling our solutions together we will also be able to provide a much higher quality solution to the California Healthcare Foundation instead of them just having access to the top few. We also look forward to future cooperative efforts for Kaggle and we will be updating this blog periodically.\r\n\r\nPlease contact us with any questions.\r\n\r\n[1]: http://www.eaglobal.org\r\n[2]: http://www.effectivealtruism.org/\r\n[3]: http://www.thelifeyoucansave.org/\r\n[4]: http://www.givewell.org/charities/top-charities", "votes": null }, { "id": "87153", "postDate": "07/27/2015 12:02:12", "content": "I suggest you try experimenting on open competitions (where the team deadline has not passed), without disrupting the competitive spirit too much.
\n\nI think you should be very careful with requiring subjects to release their predictions without a model, and realize that you are doing such during a running competition. The replication guarantee is only when the code is released later on, and happens to work to reproduce the same solution file. Furthermore competitors could also ensemble these publicly released files (though that would probably make them feel bad).
\n\nHonestly it sounds like circumventing the rules to get solutions from many individual competitors without having to team up or pass first submission deadline. On Kaggle this usually happens for nefarious reasons, that this is for a good cause, does not change that this is beyond the current rules (cheating). Competitors who do not care for "effective altruism" are placed at a disadvantage. Again, it is well known to Kaggle that using multiple accounts can benefit in ensemble building.
\n\nAre you still interested in solution and code files after the deadline has passed? Why not do that? If you want to show that "effective altruism" works to beat out competitors during a running competition, then you have to play fully by the rules. Kaggle has no team limit. Forum sharing is already working.
\n\nKaggle gets its top notch results, because competitors want to win money and fame. Because competitors want to beat others. Your experiments on running competitions, however noble and for a good cause, may damage that spirit.
", "rawMarkdown": "I suggest you try experimenting on open competitions (where the team deadline has not passed), without disrupting the competitive spirit too much.\r\n\r\nI think you should be very careful with requiring subjects to release their predictions without a model, and realize that you are doing such during a running competition. The replication guarantee is only when the code is released later on, and happens to work to reproduce the same solution file. Furthermore competitors could also ensemble these publicly released files (though that would probably make them feel bad).\r\n\r\nHonestly it sounds like circumventing the rules to get solutions from many individual competitors without having to team up or pass first submission deadline. On Kaggle this usually happens for nefarious reasons, that this is for a good cause, does not change that this is beyond the current rules (cheating). Competitors who do not care for \"effective altruism\" are placed at a disadvantage. Again, it is well known to Kaggle that using multiple accounts can benefit in ensemble building.\r\n\r\nAre you still interested in solution and code files after the deadline has passed? Why not do that? If you want to show that \"effective altruism\" works to beat out competitors during a running competition, then you have to play fully by the rules. Kaggle has no team limit. Forum sharing is already working.\r\n\r\nKaggle gets its top notch results, because competitors want to win money and fame. Because competitors want to beat others. Your experiments on running competitions, however noble and for a good cause, may damage that spirit.", "votes": null }, { "id": "87164", "postDate": "07/27/2015 13:41:28", "content": "[quote=Triskelion;87153]\nAre you still interested in solution and code files after the deadline has passed? Why not do that?\n[/quote]
\n\n+1
", "rawMarkdown": "[quote=Triskelion;87153]\r\nAre you still interested in solution and code files after the deadline has passed? Why not do that?\r\n[/quote]\r\n\r\n+1", "votes": null }, { "id": "87219", "postDate": "07/27/2015 21:58:16", "content": "Good news, it looks like there's no risk of coming close to winning, so we can forget about donation or open sourcing, but I'm still interested in how diverse the solutions are. It's unclear whether we can continue to test after the contest, and it will also be too easy to overfit to the data then (there is a certain purity in the public/private data split now), so I feel there'd still be value in taking a look at a number of training/testing outputs for curiosities sake before the contest ends.
", "rawMarkdown": "Good news, it looks like there's no risk of coming close to winning, so we can forget about donation or open sourcing, but I'm still interested in how diverse the solutions are. It's unclear whether we can continue to test after the contest, and it will also be too easy to overfit to the data then (there is a certain purity in the public/private data split now), so I feel there'd still be value in taking a look at a number of training/testing outputs for curiosities sake before the contest ends.", "votes": null } ], "comments": [ { "id": 87153, "author_name": "triskelion", "author_url": "", "post_date": "07/27/2015 12:02:12", "content": "I suggest you try experimenting on open competitions (where the team deadline has not passed), without disrupting the competitive spirit too much.
\n\nI think you should be very careful with requiring subjects to release their predictions without a model, and realize that you are doing such during a running competition. The replication guarantee is only when the code is released later on, and happens to work to reproduce the same solution file. Furthermore competitors could also ensemble these publicly released files (though that would probably make them feel bad).
\n\nHonestly it sounds like circumventing the rules to get solutions from many individual competitors without having to team up or pass first submission deadline. On Kaggle this usually happens for nefarious reasons, that this is for a good cause, does not change that this is beyond the current rules (cheating). Competitors who do not care for "effective altruism" are placed at a disadvantage. Again, it is well known to Kaggle that using multiple accounts can benefit in ensemble building.
\n\nAre you still interested in solution and code files after the deadline has passed? Why not do that? If you want to show that "effective altruism" works to beat out competitors during a running competition, then you have to play fully by the rules. Kaggle has no team limit. Forum sharing is already working.
\n\nKaggle gets its top notch results, because competitors want to win money and fame. Because competitors want to beat others. Your experiments on running competitions, however noble and for a good cause, may damage that spirit.
", "votes": null, "replies": [] }, { "id": 87164, "author_name": "inversion", "author_url": "", "post_date": "07/27/2015 13:41:28", "content": "[quote=Triskelion;87153]\nAre you still interested in solution and code files after the deadline has passed? Why not do that?\n[/quote]
\n\n+1
", "votes": null, "replies": [] }, { "id": 87219, "author_name": "nphardly", "author_url": "", "post_date": "07/27/2015 21:58:16", "content": "Good news, it looks like there's no risk of coming close to winning, so we can forget about donation or open sourcing, but I'm still interested in how diverse the solutions are. It's unclear whether we can continue to test after the contest, and it will also be too easy to overfit to the data then (there is a certain purity in the public/private data split now), so I feel there'd still be value in taking a look at a number of training/testing outputs for curiosities sake before the contest ends.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "87132": "Dear fellow Kaggle Competitor,\r\n\r\nWant to have your solution count and at the same time help save lives by contributing to effective altruism??\r\n\r\nIf so, we’d like to invite you to join a cooperative effort using ensemble methods for the Diabetic Retinopathy competition ending on Monday and the Avito Context Ad Clicks context ending on Tuesday and help win a prize that would be donated to Charity. Ensemble methods have proven time and again to win machine learning competitions and by pooling as many teams as possible, we greatly increase our chances of winning. If we do win, it will be announced at the [Effective Altruism Global][1] summit next week, and proper credit will be given to everyone who contributes.\r\n\r\nKaggle’s rules on sharing code and data privately are strict. However they do allow for all such data to be shared publicly and made available to all contestants. With this in mind, if you are interested please pay attention to the following:\r\n\r\n 1. As soon as you are able, share the outputs of both the training and testing sets on your github (or I can provide somewhere to share this) with a LICENSE file saying “Permission is granted to use this in a competition only if the winnings are donated entirely to charity”.\r\n 2. If your results are useful then, before Monday’s deadline, you will open source your code with an MIT license (without any restriction).\r\n\r\nWe encourage you to also read up on [Effective Altruism][2], “a philosophy and social movement that applies evidence and reason to determine the most effective ways to improve the world”. Any prize money won will be donated to charities recommended by [The Life You Can Save][3] and [Give Well][4]. We are happy to hear about any other requests or suggestions.\r\n\r\nAs a last comment, by pooling our solutions together we will also be able to provide a much higher quality solution to the California Healthcare Foundation instead of them just having access to the top few. We also look forward to future cooperative efforts for Kaggle and we will be updating this blog periodically.\r\n\r\nPlease contact us with any questions.\r\n\r\n[1]: http://www.eaglobal.org\r\n[2]: http://www.effectivealtruism.org/\r\n[3]: http://www.thelifeyoucansave.org/\r\n[4]: http://www.givewell.org/charities/top-charities", "87153": "I suggest you try experimenting on open competitions (where the team deadline has not passed), without disrupting the competitive spirit too much.\r\n\r\nI think you should be very careful with requiring subjects to release their predictions without a model, and realize that you are doing such during a running competition. The replication guarantee is only when the code is released later on, and happens to work to reproduce the same solution file. Furthermore competitors could also ensemble these publicly released files (though that would probably make them feel bad).\r\n\r\nHonestly it sounds like circumventing the rules to get solutions from many individual competitors without having to team up or pass first submission deadline. On Kaggle this usually happens for nefarious reasons, that this is for a good cause, does not change that this is beyond the current rules (cheating). Competitors who do not care for \"effective altruism\" are placed at a disadvantage. Again, it is well known to Kaggle that using multiple accounts can benefit in ensemble building.\r\n\r\nAre you still interested in solution and code files after the deadline has passed? Why not do that? If you want to show that \"effective altruism\" works to beat out competitors during a running competition, then you have to play fully by the rules. Kaggle has no team limit. Forum sharing is already working.\r\n\r\nKaggle gets its top notch results, because competitors want to win money and fame. Because competitors want to beat others. Your experiments on running competitions, however noble and for a good cause, may damage that spirit.", "87164": "[quote=Triskelion;87153]\r\nAre you still interested in solution and code files after the deadline has passed? Why not do that?\r\n[/quote]\r\n\r\n+1", "87219": "Good news, it looks like there's no risk of coming close to winning, so we can forget about donation or open sourcing, but I'm still interested in how diverse the solutions are. It's unclear whether we can continue to test after the contest, and it will also be too easy to overfit to the data then (there is a certain purity in the public/private data split now), so I feel there'd still be value in taking a look at a number of training/testing outputs for curiosities sake before the contest ends." }, "source": "meta" }