While stage one is just for practice, cheating during this phase discourages newcomers who don't understand that the leaderboard is not \"solved\". You are cheating if you intentionally or accidentally use the game outcomes in your submission. When forecasting, your algorithm can't look ahead in time!
\n\nWe just removed any submissions with a score < 0.2 and will continue to periodically knock out unreasonably good submissions (and will adjust this threshold depending on what we observe... even 0.2 is very generous compared with historical scores).
", "messages": [ { "id": 749377, "postDate": "2020-02-18T16:33:10.803Z", "content": "While stage one is just for practice, cheating during this phase discourages newcomers who don't understand that the leaderboard is not \"solved\". You are cheating if you intentionally or accidentally use the game outcomes in your submission. When forecasting, your algorithm can't look ahead in time!
\n\nWe just removed any submissions with a score < 0.2 and will continue to periodically knock out unreasonably good submissions (and will adjust this threshold depending on what we observe... even 0.2 is very generous compared with historical scores).
", "rawMarkdown": "While stage one is just for practice, cheating during this phase discourages newcomers who don't understand that the leaderboard is not \"solved\". You are cheating if you intentionally or accidentally use the game outcomes in your submission. When forecasting, your algorithm can't look ahead in time!\n\nWe just removed any submissions with a score < 0.2 and will continue to periodically knock out unreasonably good submissions (and will adjust this threshold depending on what we observe... even 0.2 is very generous compared with historical scores).", "votes": 23 }, { "id": 751947, "postDate": "2020-02-20T17:00:33.460Z", "content": "I am a newcomer and was guilty of this accidentally. Is this an example of data leakage? I want to understand exactly what is going on. When making predictions for a given season (e.g. 2015), should one use a model that did not include any training examples from just that season (2015), or must you hold out all seasons after that as well (>=2015, because it hasn't happened yet, even though they may be independent)? Any response is much appreciated, thank you.
", "rawMarkdown": "I am a newcomer and was guilty of this accidentally. Is this an example of data leakage? I want to understand exactly what is going on. When making predictions for a given season (e.g. 2015), should one use a model that did not include any training examples from just that season (2015), or must you hold out all seasons after that as well (>=2015, because it hasn't happened yet, even though they may be independent)? Any response is much appreciated, thank you.", "votes": 1, "replies": [ { "id": 753308, "postDate": "2020-02-22T02:25:09.087Z", "content": "As far as I understand, as long as you are using the past data to predict the future outcome, you are on the safe side. Many public kernels including mine simply use Kfold for all the data, which generates cases where future data are trained to predict the past, causing the leak.
", "rawMarkdown": "As far as I understand, as long as you are using the past data to predict the future outcome, you are on the safe side. Many public kernels including mine simply use Kfold for all the data, which generates cases where future data are trained to predict the past, causing the leak." } ] }, { "id": 749471, "postDate": "2020-02-18T18:14:37.880Z", "content": "Well I'm sure that's not what the top 2 teams have done :)
", "rawMarkdown": "Well I'm sure that's not what the top 2 teams have done :)", "votes": 1 }, { "id": 761731, "postDate": "2020-03-02T22:06:55.967Z", "content": "Wouldn't the judgment \"because the results of past competitions are like this\" prevent new discoveries?\nCertainly, there were many suspicious submissions.\nIt was doubtful that the same numbers were lined up.\nIn many cases, 0.0 was obviously a copy of past data.\nHowever, is there a basis for uniformly deleting?\nIsn't this a bit rude judgment?
\n\n”過去のコンペティションでの結果がこうだから”という判断は新しい発見を妨げるものではないでしょうか？\n確かに怪しいサブミットはたくさんありました。\n同じ数字が並んでいるのも怪しかった。\n0.0なんて明らかに過去のデータをコピーしたものだろうというものもたくさんありました。\nしかし、一律で削除というのは根拠があるのでしょうか？\n少々横暴な判断ではないでしょうか？
", "rawMarkdown": "Wouldn't the judgment \"because the results of past competitions are like this\" prevent new discoveries?\nCertainly, there were many suspicious submissions.\nIt was doubtful that the same numbers were lined up.\nIn many cases, 0.0 was obviously a copy of past data.\nHowever, is there a basis for uniformly deleting?\nIsn't this a bit rude judgment?\n\n\n”過去のコンペティションでの結果がこうだから”という判断は新しい発見を妨げるものではないでしょうか？\n確かに怪しいサブミットはたくさんありました。\n同じ数字が並んでいるのも怪しかった。\n0.0なんて明らかに過去のデータをコピーしたものだろうというものもたくさんありました。\nしかし、一律で削除というのは根拠があるのでしょうか？\n少々横暴な判断ではないでしょうか？", "votes": -1, "replies": [ { "id": 761816, "postDate": "2020-03-03T00:33:10.140Z", "content": "If your score isn't from cheating and getting 0.2-, you may likely to get top at private LB ( Last result )\nfinal result is from private,\nthen if you're confident you don't need to stick public LB,\nI'm sure there're lots of teams cheating for public LB now,\n(Last year, too, \nAs you know when there's a leak kagglers will find them to pretend like they're modeling well)\nthen it's no use submitting to public LB,\nbut fortunately, no any cheating is available for stage2, last results(because it's pure future prediction)\nwe can trust our validation and only submit them at second stage.\nI'm not submitting to stage1, neither😂
", "rawMarkdown": "If your score isn't from cheating and getting 0.2-, you may likely to get top at private LB ( Last result )\nfinal result is from private,\nthen if you're confident you don't need to stick public LB,\nI'm sure there're lots of teams cheating for public LB now,\n(Last year, too, \nAs you know when there's a leak kagglers will find them to pretend like they're modeling well)\nthen it's no use submitting to public LB,\nbut fortunately, no any cheating is available for stage2, last results(because it's pure future prediction)\nwe can trust our validation and only submit them at second stage.\nI'm not submitting to stage1, neither😂" } ] }, { "id": 761726, "postDate": "2020-03-02T21:52:21.997Z", "content": "私のデータが削除されましたが、過去の結果は一切使用していません。\n納得がいかない。\n一律で0.2未満を削除するという根拠を示していただきたい。\n私は過去のデータを使用していないという根拠は示せます。
", "rawMarkdown": "私のデータが削除されましたが、過去の結果は一切使用していません。\n納得がいかない。\n一律で0.2未満を削除するという根拠を示していただきたい。\n私は過去のデータを使用していないという根拠は示せます。", "votes": -1 }, { "id": 760282, "postDate": "2020-03-01T04:20:24.240Z", "content": "過去の結果など利用していないが、普通に0.2未満が出たんですが..\n問題あるんですか？
", "rawMarkdown": "過去の結果など利用していないが、普通に0.2未満が出たんですが..\n問題あるんですか？", "votes": -1 }, { "id": 757935, "postDate": "2020-02-27T09:09:46.277Z", "content": "ルールを理解できていませんでした。すいません。
", "rawMarkdown": "ルールを理解できていませんでした。すいません。", "votes": -1 }, { "id": 750364, "postDate": "2020-02-19T10:20:41.593Z", "content": "This appears not to be my account, I need to confirm with my email: robertaustine4@gmail.com please!
", "rawMarkdown": "This appears not to be my account, I need to confirm with my email: robertaustine4@gmail.com please!", "votes": -1 }, { "id": 757755, "postDate": "2020-02-27T04:29:46.640Z", "content": "Sorry, I didn't notice this discussion. I suggest we put this on the overview so that people don't make the same mistake as me.
", "rawMarkdown": "Sorry, I didn't notice this discussion. I suggest we put this on the overview so that people don't make the same mistake as me. " }, { "id": 755890, "postDate": "2020-02-25T08:29:28.033Z", "content": "agree
", "rawMarkdown": "agree" }, { "id": 753296, "postDate": "2020-02-22T01:47:16.393Z", "content": "I really want to say this, too\nI want to check my purely no leak model on Public LB but seems lot's of people're using leakage in order to pretend just Public LB looks like \"strong kaggler\",\nalthough all of them know that we can't do any cheating at private LB for this competition.\n(Of course there's possibility that all of top Public LB teams are not using leak but it'll be easily detected when they're not high LB at private LB of this year)
", "rawMarkdown": "I really want to say this, too\nI want to check my purely no leak model on Public LB but seems lot's of people're using leakage in order to pretend just Public LB looks like \"strong kaggler\",\nalthough all of them know that we can't do any cheating at private LB for this competition.\n(Of course there's possibility that all of top Public LB teams are not using leak but it'll be easily detected when they're not high LB at private LB of this year)" }, { "id": 752479, "postDate": "2020-02-21T04:07:34.073Z", "content": "My best score was <0.20 with a simple LGB and it did not appear to be a leak (very same kernel scored ~0.3 in mens' data). But could it be in womens' case somehow??
", "rawMarkdown": "My best score was <0.20 with a simple LGB and it did not appear to be a leak (very same kernel scored ~0.3 in mens' data). But could it be in womens' case somehow??", "replies": [ { "id": 753105, "postDate": "2020-02-21T18:03:35.297Z", "content": "You are probably using data from previous years' March Madness Tournaments while training your model. A score that low is highly unlikely without it. See this thread:
\n\n", "rawMarkdown": "You are probably using data from previous years' March Madness Tournaments while training your model. A score that low is highly unlikely without it. See this thread:\n\nhttps://www.kaggle.com/c/google-cloud-ncaa-march-madness-2020-division-1-mens-tournament/discussion/131028\n", "votes": 2 }, { "id": 753235, "postDate": "2020-02-21T22:29:24.920Z", "content": "Right...i’ve Got to be into this...thanks for your reply!
", "rawMarkdown": "Right...i’ve Got to be into this...thanks for your reply!" } ] }, { "id": 752318, "postDate": "2020-02-20T21:45:30.330Z", "content": "yes!
", "rawMarkdown": "yes!" }, { "id": 752224, "postDate": "2020-02-20T19:51:30.433Z", "content": "agree
", "rawMarkdown": "agree" }, { "id": 749641, "postDate": "2020-02-18T20:35:25.497Z", "content": "i am totally agree with you
", "rawMarkdown": "i am totally agree with you " }, { "id": 753855, "postDate": "2020-02-22T18:16:51.687Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 753895, "postDate": "2020-02-22T19:37:00.337Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 751947, "author_name": "Ryan Marshall", "author_url": "", "post_date": "2020-02-20T17:00:33.460000", "content": "I am a newcomer and was guilty of this accidentally. Is this an example of data leakage? I want to understand exactly what is going on. When making predictions for a given season (e.g. 2015), should one use a model that did not include any training examples from just that season (2015), or must you hold out all seasons after that as well (>=2015, because it hasn't happened yet, even though they may be independent)? Any response is much appreciated, thank you.
", "votes": 1, "replies": [ { "id": 753308, "author_name": "katsu1110", "author_url": "", "post_date": "2020-02-22T02:25:09.087000", "content": "As far as I understand, as long as you are using the past data to predict the future outcome, you are on the safe side. Many public kernels including mine simply use Kfold for all the data, which generates cases where future data are trained to predict the past, causing the leak.
", "votes": 0, "replies": [] } ] }, { "id": 749471, "author_name": "Marco Gorelli", "author_url": "", "post_date": "2020-02-18T18:14:37.880000", "content": "Well I'm sure that's not what the top 2 teams have done :)
", "votes": 1, "replies": [] }, { "id": 761731, "author_name": "HideShoyama", "author_url": "", "post_date": "2020-03-02T22:06:55.967000", "content": "Wouldn't the judgment \"because the results of past competitions are like this\" prevent new discoveries?\nCertainly, there were many suspicious submissions.\nIt was doubtful that the same numbers were lined up.\nIn many cases, 0.0 was obviously a copy of past data.\nHowever, is there a basis for uniformly deleting?\nIsn't this a bit rude judgment?
\n\n”過去のコンペティションでの結果がこうだから”という判断は新しい発見を妨げるものではないでしょうか？\n確かに怪しいサブミットはたくさんありました。\n同じ数字が並んでいるのも怪しかった。\n0.0なんて明らかに過去のデータをコピーしたものだろうというものもたくさんありました。\nしかし、一律で削除というのは根拠があるのでしょうか？\n少々横暴な判断ではないでしょうか？
", "votes": -1, "replies": [ { "id": 761816, "author_name": "Ryunosuke Ishizaki", "author_url": "", "post_date": "2020-03-03T00:33:10.140000", "content": "If your score isn't from cheating and getting 0.2-, you may likely to get top at private LB ( Last result )\nfinal result is from private,\nthen if you're confident you don't need to stick public LB,\nI'm sure there're lots of teams cheating for public LB now,\n(Last year, too, \nAs you know when there's a leak kagglers will find them to pretend like they're modeling well)\nthen it's no use submitting to public LB,\nbut fortunately, no any cheating is available for stage2, last results(because it's pure future prediction)\nwe can trust our validation and only submit them at second stage.\nI'm not submitting to stage1, neither😂
", "votes": 0, "replies": [] } ] }, { "id": 761726, "author_name": "HideShoyama", "author_url": "", "post_date": "2020-03-02T21:52:21.997000", "content": "私のデータが削除されましたが、過去の結果は一切使用していません。\n納得がいかない。\n一律で0.2未満を削除するという根拠を示していただきたい。\n私は過去のデータを使用していないという根拠は示せます。
", "votes": -1, "replies": [] }, { "id": 760282, "author_name": "HideShoyama", "author_url": "", "post_date": "2020-03-01T04:20:24.240000", "content": "過去の結果など利用していないが、普通に0.2未満が出たんですが..\n問題あるんですか？
", "votes": -1, "replies": [] }, { "id": 757935, "author_name": "tensor choko", "author_url": "", "post_date": "2020-02-27T09:09:46.277000", "content": "ルールを理解できていませんでした。すいません。
", "votes": -1, "replies": [] }, { "id": 750364, "author_name": "robert austine", "author_url": "", "post_date": "2020-02-19T10:20:41.593000", "content": "This appears not to be my account, I need to confirm with my email: robertaustine4@gmail.com please!
", "votes": -1, "replies": [] }, { "id": 757755, "author_name": "Hao-Hsiang Wu", "author_url": "", "post_date": "2020-02-27T04:29:46.640000", "content": "Sorry, I didn't notice this discussion. I suggest we put this on the overview so that people don't make the same mistake as me.
", "votes": 0, "replies": [] }, { "id": 755890, "author_name": "Alex", "author_url": "", "post_date": "2020-02-25T08:29:28.033000", "content": "agree
", "votes": 0, "replies": [] }, { "id": 753296, "author_name": "Ryunosuke Ishizaki", "author_url": "", "post_date": "2020-02-22T01:47:16.393000", "content": "I really want to say this, too\nI want to check my purely no leak model on Public LB but seems lot's of people're using leakage in order to pretend just Public LB looks like \"strong kaggler\",\nalthough all of them know that we can't do any cheating at private LB for this competition.\n(Of course there's possibility that all of top Public LB teams are not using leak but it'll be easily detected when they're not high LB at private LB of this year)
", "votes": 0, "replies": [] }, { "id": 752479, "author_name": "katsu1110", "author_url": "", "post_date": "2020-02-21T04:07:34.073000", "content": "My best score was <0.20 with a simple LGB and it did not appear to be a leak (very same kernel scored ~0.3 in mens' data). But could it be in womens' case somehow??
", "votes": 0, "replies": [ { "id": 753105, "author_name": "KrisztianSz", "author_url": "", "post_date": "2020-02-21T18:03:35.297000", "content": "You are probably using data from previous years' March Madness Tournaments while training your model. A score that low is highly unlikely without it. See this thread:
\n\n", "votes": 2, "replies": [] }, { "id": 753235, "author_name": "katsu1110", "author_url": "", "post_date": "2020-02-21T22:29:24.920000", "content": "Right...i’ve Got to be into this...thanks for your reply!
", "votes": 0, "replies": [] } ] }, { "id": 752318, "author_name": "", "author_url": "", "post_date": "2020-02-20T21:45:30.330000", "content": "yes!
", "votes": 0, "replies": [] }, { "id": 752224, "author_name": "", "author_url": "", "post_date": "2020-02-20T19:51:30.433000", "content": "agree
", "votes": 0, "replies": [] }, { "id": 749641, "author_name": "Mina Nabil", "author_url": "", "post_date": "2020-02-18T20:35:25.497000", "content": "i am totally agree with you
", "votes": 0, "replies": [] }, { "id": 753855, "author_name": "", "author_url": "", "post_date": "2020-02-22T18:16:51.687000", "content": "", "votes": 0, "replies": [ { "id": 753895, "author_name": "", "author_url": "", "post_date": "2020-02-22T19:37:00.337000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "749377": "While stage one is just for practice, cheating during this phase discourages newcomers who don't understand that the leaderboard is not \"solved\". You are cheating if you intentionally or accidentally use the game outcomes in your submission. When forecasting, your algorithm can't look ahead in time!\n\nWe just removed any submissions with a score < 0.2 and will continue to periodically knock out unreasonably good submissions (and will adjust this threshold depending on what we observe... even 0.2 is very generous compared with historical scores).", "751947": "I am a newcomer and was guilty of this accidentally. Is this an example of data leakage? I want to understand exactly what is going on. When making predictions for a given season (e.g. 2015), should one use a model that did not include any training examples from just that season (2015), or must you hold out all seasons after that as well (>=2015, because it hasn't happened yet, even though they may be independent)? Any response is much appreciated, thank you.", "749471": "Well I'm sure that's not what the top 2 teams have done :)", "761731": "Wouldn't the judgment \"because the results of past competitions are like this\" prevent new discoveries?\nCertainly, there were many suspicious submissions.\nIt was doubtful that the same numbers were lined up.\nIn many cases, 0.0 was obviously a copy of past data.\nHowever, is there a basis for uniformly deleting?\nIsn't this a bit rude judgment?\n\n\n”過去のコンペティションでの結果がこうだから”という判断は新しい発見を妨げるものではないでしょうか？\n確かに怪しいサブミットはたくさんありました。\n同じ数字が並んでいるのも怪しかった。\n0.0なんて明らかに過去のデータをコピーしたものだろうというものもたくさんありました。\nしかし、一律で削除というのは根拠があるのでしょうか？\n少々横暴な判断ではないでしょうか？", "761726": "私のデータが削除されましたが、過去の結果は一切使用していません。\n納得がいかない。\n一律で0.2未満を削除するという根拠を示していただきたい。\n私は過去のデータを使用していないという根拠は示せます。", "760282": "過去の結果など利用していないが、普通に0.2未満が出たんですが..\n問題あるんですか？", "757935": "ルールを理解できていませんでした。すいません。", "750364": "This appears not to be my account, I need to confirm with my email: robertaustine4@gmail.com please!", "757755": "Sorry, I didn't notice this discussion. I suggest we put this on the overview so that people don't make the same mistake as me. ", "755890": "agree", "753296": "I really want to say this, too\nI want to check my purely no leak model on Public LB but seems lot's of people're using leakage in order to pretend just Public LB looks like \"strong kaggler\",\nalthough all of them know that we can't do any cheating at private LB for this competition.\n(Of course there's possibility that all of top Public LB teams are not using leak but it'll be easily detected when they're not high LB at private LB of this year)", "752479": "My best score was <0.20 with a simple LGB and it did not appear to be a leak (very same kernel scored ~0.3 in mens' data). But could it be in womens' case somehow??", "752318": "yes!", "752224": "agree", "749641": "i am totally agree with you ", "753855": "" } }