I have been closely following the \"Dance with Ensemble\" team (currently Rank #1 on the public Leaderboard). The score is way ahead of the second position score (unlike other rankings which are quite close enough). And this team continuously keep improving beating their own score.
\n\nWhile it will be really interesting to know the team's approach post this competition. But I was just thinking if this is a case of over fitting?
\n\nNote: Please don't consider this post as a point to demotivate anyone. Just curious to know how do we ensure that we avoid over fitting.
", "messages": [ { "id": 349063, "postDate": "2018-06-27T18:19:46.690Z", "content": "We are probably as overfitting as every other top team I would say. :) But yeah, we did make sure that we have consistent cv/lb performance along the way. And the rest is things that we cannot really control. We will see in a few hours. But thanks for raising the question, I feel honoured. :)
", "rawMarkdown": "We are probably as overfitting as every other top team I would say. :) But yeah, we did make sure that we have consistent cv/lb performance along the way. And the rest is things that we cannot really control. We will see in a few hours. But thanks for raising the question, I feel honoured. :)", "votes": 13 }, { "id": 348845, "postDate": "2018-06-27T11:40:32.753Z", "content": "Sorry guys but I don't see the need of throwing out this kind of suspicion about if a particular team is overfitting or not, and this even before the end of the competition, apart from minimising people's hard work (and talent) .
\n\nAnyway , members of that particular team are talented kagglers , who were at the top of LB since the beginning as solo competitors, so even the score is quit impressive , I don't know why they would \"definitely likely\" overfitting ( as you said ).
\n\nBut, as many here I can't wait to see their approach for such huge domination :)
", "rawMarkdown": "Sorry guys but I don't see the need of throwing out this kind of suspicion about if a particular team is overfitting or not, and this even before the end of the competition, apart from minimising people's hard work (and talent) .\n\nAnyway , members of that particular team are talented kagglers , who were at the top of LB since the beginning as solo competitors, so even the score is quit impressive , I don't know why they would \"definitely likely\" overfitting ( as you said ). \n\n But, as many here I can't wait to see their approach for such huge domination :)\n\n", "votes": 11, "replies": [ { "id": 348861, "postDate": "2018-06-27T12:09:51.037Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 348908, "postDate": "2018-06-27T14:04:11.800Z", "content": "Of course all models are wrong....\nBut some are useful for winning Kaggle competitions.
", "rawMarkdown": "Of course all models are wrong....\nBut some are useful for winning Kaggle competitions.", "votes": 6 }, { "id": 348877, "postDate": "2018-06-27T12:45:11.140Z", "content": "I'd say it's overwhelmingly more likely that everyone else is underfitting ;)
", "rawMarkdown": "I'd say it's overwhelmingly more likely that everyone else is underfitting ;)", "votes": 4, "replies": [ { "id": 348909, "postDate": "2018-06-27T14:05:11.453Z", "content": "Nice to see some good humor. Some people took it really seriously. May be i should have kept the question more general.
", "rawMarkdown": "Nice to see some good humor. Some people took it really seriously. May be i should have kept the question more general." } ] }, { "id": 349050, "postDate": "2018-06-27T17:54:45.903Z", "content": "There are so many avenues to go with this data... I feel like we could have big gaps in areas covered, so not surprised to see top team well ahead. Looking at the lb progression over the last weeks, if the comp went another month we'd probably have 10 teams sub 0.21...
", "rawMarkdown": "There are so many avenues to go with this data... I feel like we could have big gaps in areas covered, so not surprised to see top team well ahead. Looking at the lb progression over the last weeks, if the comp went another month we'd probably have 10 teams sub 0.21...", "votes": 3 }, { "id": 348820, "postDate": "2018-06-27T11:03:08.663Z", "content": "I have been closely following the \"Dance with Ensemble\" team (currently Rank #1 on the public Leaderboard). The score is way ahead of the second position score (unlike other rankings which are quite close enough). And this team continuously keep improving beating their own score.
\n\nWhile it will be really interesting to know the team's approach post this competition. But I was just thinking if this is a case of over fitting?
\n\nNote: Please don't consider this post as a point to demotivate anyone. Just curious to know how do we ensure that we avoid over fitting.
", "rawMarkdown": "I have been closely following the \"Dance with Ensemble\" team (currently Rank #1 on the public Leaderboard). The score is way ahead of the second position score (unlike other rankings which are quite close enough). And this team continuously keep improving beating their own score. \n\nWhile it will be really interesting to know the team's approach post this competition. But I was just thinking if this is a case of over fitting?\n\nNote: Please don't consider this post as a point to demotivate anyone. Just curious to know how do we ensure that we avoid over fitting.", "votes": -2 }, { "id": 348837, "postDate": "2018-06-27T11:20:34.213Z", "content": "Who doesn’t? My team chooses to ignore it partly because of not having time and/or ignorance.
", "rawMarkdown": "Who doesn’t? My team chooses to ignore it partly because of not having time and/or ignorance. " }, { "id": 348832, "postDate": "2018-06-27T11:12:55.073Z", "rawMarkdown": "", "votes": 2, "isDeleted": true, "replies": [ { "id": 348834, "postDate": "2018-06-27T11:18:27.130Z", "content": "“Definitely likely” :)
", "rawMarkdown": "“Definitely likely” :) ", "votes": 4 } ] } ], "comments": [ { "id": 349063, "author_name": "Little Boat", "author_url": "", "post_date": "2018-06-27T18:19:46.690000", "content": "We are probably as overfitting as every other top team I would say. :) But yeah, we did make sure that we have consistent cv/lb performance along the way. And the rest is things that we cannot really control. We will see in a few hours. But thanks for raising the question, I feel honoured. :)
", "votes": 13, "replies": [] }, { "id": 348845, "author_name": "Serigne ", "author_url": "", "post_date": "2018-06-27T11:40:32.753000", "content": "Sorry guys but I don't see the need of throwing out this kind of suspicion about if a particular team is overfitting or not, and this even before the end of the competition, apart from minimising people's hard work (and talent) .
\n\nAnyway , members of that particular team are talented kagglers , who were at the top of LB since the beginning as solo competitors, so even the score is quit impressive , I don't know why they would \"definitely likely\" overfitting ( as you said ).
\n\nBut, as many here I can't wait to see their approach for such huge domination :)
", "votes": 11, "replies": [ { "id": 348861, "author_name": "", "author_url": "", "post_date": "2018-06-27T12:09:51.037000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 348908, "author_name": "pocket", "author_url": "", "post_date": "2018-06-27T14:04:11.800000", "content": "Of course all models are wrong....\nBut some are useful for winning Kaggle competitions.
", "votes": 6, "replies": [] }, { "id": 348877, "author_name": "Joe Eddy", "author_url": "", "post_date": "2018-06-27T12:45:11.140000", "content": "I'd say it's overwhelmingly more likely that everyone else is underfitting ;)
", "votes": 4, "replies": [ { "id": 348909, "author_name": "abdul0807", "author_url": "", "post_date": "2018-06-27T14:05:11.453000", "content": "Nice to see some good humor. Some people took it really seriously. May be i should have kept the question more general.
", "votes": 0, "replies": [] } ] }, { "id": 349050, "author_name": "Darragh", "author_url": "", "post_date": "2018-06-27T17:54:45.903000", "content": "There are so many avenues to go with this data... I feel like we could have big gaps in areas covered, so not surprised to see top team well ahead. Looking at the lb progression over the last weeks, if the comp went another month we'd probably have 10 teams sub 0.21...
", "votes": 3, "replies": [] }, { "id": 348837, "author_name": "yimacs", "author_url": "", "post_date": "2018-06-27T11:20:34.213000", "content": "Who doesn’t? My team chooses to ignore it partly because of not having time and/or ignorance.
", "votes": 0, "replies": [] }, { "id": 348832, "author_name": "", "author_url": "", "post_date": "2018-06-27T11:12:55.073000", "content": "", "votes": 2, "replies": [ { "id": 348834, "author_name": "yimacs", "author_url": "", "post_date": "2018-06-27T11:18:27.130000", "content": "“Definitely likely” :)
", "votes": 4, "replies": [] } ] } ], "raw_markdown_by_id": { "349063": "We are probably as overfitting as every other top team I would say. :) But yeah, we did make sure that we have consistent cv/lb performance along the way. And the rest is things that we cannot really control. We will see in a few hours. But thanks for raising the question, I feel honoured. :)", "348845": "Sorry guys but I don't see the need of throwing out this kind of suspicion about if a particular team is overfitting or not, and this even before the end of the competition, apart from minimising people's hard work (and talent) .\n\nAnyway , members of that particular team are talented kagglers , who were at the top of LB since the beginning as solo competitors, so even the score is quit impressive , I don't know why they would \"definitely likely\" overfitting ( as you said ). \n\n But, as many here I can't wait to see their approach for such huge domination :)\n\n", "348908": "Of course all models are wrong....\nBut some are useful for winning Kaggle competitions.", "348877": "I'd say it's overwhelmingly more likely that everyone else is underfitting ;)", "349050": "There are so many avenues to go with this data... I feel like we could have big gaps in areas covered, so not surprised to see top team well ahead. Looking at the lb progression over the last weeks, if the comp went another month we'd probably have 10 teams sub 0.21...", "348820": "I have been closely following the \"Dance with Ensemble\" team (currently Rank #1 on the public Leaderboard). The score is way ahead of the second position score (unlike other rankings which are quite close enough). And this team continuously keep improving beating their own score. \n\nWhile it will be really interesting to know the team's approach post this competition. But I was just thinking if this is a case of over fitting?\n\nNote: Please don't consider this post as a point to demotivate anyone. Just curious to know how do we ensure that we avoid over fitting.", "348837": "Who doesn’t? My team chooses to ignore it partly because of not having time and/or ignorance. ", "348832": "" } }