As many, I wondered if top score was not due to some leak. I now don't think there is a leak because my first sub at LB 0.931 is just a 5 folds single model without any leak nor external data.
\nI am sharing not to brag, but rather to give hope to all those who could be discouraged by the incredible top score.
", "messages": [ { "id": 1159884, "postDate": "2021-01-19T14:24:22.237Z", "content": "As many, I wondered if top score was not due to some leak. I now don't think there is a leak because my first sub at LB 0.931 is just a 5 folds single model without any leak nor external data.
\nI am sharing not to brag, but rather to give hope to all those who could be discouraged by the incredible top score.
", "rawMarkdown": "As many, I wondered if top score was not due to some leak. I now don't think there is a leak because my first sub at LB 0.931 is just a 5 folds single model without any leak nor external data. \n\nI am sharing not to brag, but rather to give hope to all those who could be discouraged by the incredible top score.", "votes": 66 }, { "id": 1160012, "postDate": "2021-01-19T15:58:48.443Z", "content": "The fact that you were about to reach 0.931 in one submission without using any leak is already a lot of information disclosed ;)
", "rawMarkdown": "The fact that you were about to reach 0.931 in one submission without using any leak is already a lot of information disclosed ;)", "votes": 9, "replies": [ { "id": 1160169, "postDate": "2021-01-19T17:58:24.980Z", "content": "I hope I won't regret sharing it ;)
", "rawMarkdown": "I hope I won't regret sharing it ;)", "votes": 8 } ] }, { "id": 1164109, "postDate": "2021-01-22T06:51:36.030Z", "content": "@cpmpml , I remember watching one of your talks \"Beyond Feature Engineering and HPO\" , in that you mention one user who win comp with just 1 submission may be you might be able to replicate it this time 😄😄.
", "rawMarkdown": "@cpmpml , I remember watching one of your talks \"Beyond Feature Engineering and HPO\" , in that you mention one user who win comp with just 1 submission may be you might be able to replicate it this time 😄😄.", "votes": 3, "replies": [ { "id": 1164295, "postDate": "2021-01-22T09:26:01.303Z", "content": "Indeed, the user was @speculation and he won expedia competition with one sub indeed: https://www.kaggle.com/c/expedia-hotel-recommendations/leaderboard
\nThis is much much better than me having one sub that ranked 3rd in the middle of the competition.
", "rawMarkdown": "Indeed, the user was @speculation and he won expedia competition with one sub indeed: https://www.kaggle.com/c/expedia-hotel-recommendations/leaderboard\n\nThis is much much better than me having one sub that ranked 3rd in the middle of the competition.", "votes": 4 } ] }, { "id": 1160016, "postDate": "2021-01-19T16:02:15.820Z", "content": "Did you expect this score or you got better/worse than you expected?
", "rawMarkdown": "Did you expect this score or you got better/worse than you expected?", "votes": 3, "replies": [ { "id": 1160134, "postDate": "2021-01-19T17:27:51.243Z", "content": "It could have been better, i.e. my CV is higher ;)
\nBut I am happy with what it is, it confirms that my baseline is sound.
", "rawMarkdown": "It could have been better, i.e. my CV is higher ;)\n\nBut I am happy with what it is, it confirms that my baseline is sound.", "votes": 5 }, { "id": 1183346, "postDate": "2021-02-02T23:04:53.737Z", "content": "It sounds like you are using really short segments?
", "rawMarkdown": "It sounds like you are using really short segments?" }, { "id": 1183354, "postDate": "2021-02-02T23:21:16.063Z", "content": "What makes you think that? Anyway, I'll try to not share what I do before competition end. It seems that the little I shared helped a lot of people already ;)
", "rawMarkdown": "What makes you think that? Anyway, I'll try to not share what I do before competition end. It seems that the little I shared helped a lot of people already ;)", "votes": 1 }, { "id": 1183360, "postDate": "2021-02-02T23:30:27.707Z", "content": "I only get CV +0.92 if I use shorter segments :)
", "rawMarkdown": "I only get CV +0.92 if I use shorter segments :)", "votes": 1 } ] }, { "id": 1163306, "postDate": "2021-01-21T16:28:33.603Z", "content": "May I ask what a leak here means?
\nAnyways, thanks for the encouragement!
", "rawMarkdown": "May I ask what a leak here means?\n\nAnyways, thanks for the encouragement!", "votes": 1, "replies": [ { "id": 1163361, "postDate": "2021-01-21T16:47:24.493Z", "content": "People don't always agree on what a leak is, but here is how Kaggle defines it: https://www.kaggle.com/docs/competitions#leakage
", "rawMarkdown": "People don't always agree on what a leak is, but here is how Kaggle defines it: https://www.kaggle.com/docs/competitions#leakage\n\n", "votes": 1 }, { "id": 1163438, "postDate": "2021-01-21T17:12:42.050Z", "content": "Oh I see, thank you!
", "rawMarkdown": "Oh I see, thank you!" } ] }, { "id": 1160822, "postDate": "2021-01-20T06:42:54.927Z", "content": "Very interesting, thanks for sharing. I might consider checking this competition then. :D
", "rawMarkdown": "Very interesting, thanks for sharing. I might consider checking this competition then. :D", "votes": 1 }, { "id": 1160267, "postDate": "2021-01-19T19:09:35.167Z", "content": "This competition taught me a lot so far and now I can't wait to see the winning solutions. I can't even imagine what will happen then :)
", "rawMarkdown": "This competition taught me a lot so far and now I can't wait to see the winning solutions. I can't even imagine what will happen then :)", "votes": 1 }, { "id": 1160265, "postDate": "2021-01-19T19:07:09.190Z", "content": "Just woow 🔥
\nOnly 1 fire because of the total submission count :)
", "rawMarkdown": "Just woow 🔥\n\nOnly 1 fire because of the total submission count :)", "votes": 1 }, { "id": 1160210, "postDate": "2021-01-19T18:30:20.310Z", "content": "Without no leak? Kudos to you👍
", "rawMarkdown": "Without no leak? Kudos to you👍", "votes": 1 }, { "id": 1164308, "postDate": "2021-01-22T09:37:36.117Z", "content": "I feel like the top scores have gone wild since your post :D
", "rawMarkdown": "I feel like the top scores have gone wild since your post :D", "votes": 2, "replies": [ { "id": 1168470, "postDate": "2021-01-25T01:00:21.047Z", "content": "Wait till I move to larger models than effcientnet B0 ;)
", "rawMarkdown": "Wait till I move to larger models than effcientnet B0 ;)", "votes": 4 }, { "id": 1173871, "postDate": "2021-01-28T06:20:10.960Z", "content": "I'm still waiting.
", "rawMarkdown": "I'm still waiting. ", "votes": 3 }, { "id": 1173929, "postDate": "2021-01-28T07:18:09.367Z", "content": "Yeah, smaller models seem to work better.
", "rawMarkdown": "Yeah, smaller models seem to work better." }, { "id": 1176061, "postDate": "2021-01-29T13:33:46.227Z", "content": "\n\nI'm still waiting.
\n
I spent 5 days on another (non kaggle) competition, I resumed here yesterday. I hope to get something better within next few days.
", "rawMarkdown": "> I'm still waiting. \n\nI spent 5 days on another (non kaggle) competition, I resumed here yesterday. I hope to get something better within next few days.", "votes": 3 }, { "id": 1184312, "postDate": "2021-02-03T13:37:27.913Z", "content": "For those interested, it was this competition https://www.bookingchallenge.com/final-leaderboard-results
\nI was lucky to win with @titericz , @cdeotte @jiweiliu and other NVIDIA colleagues. :D
", "rawMarkdown": "For those interested, it was this competition https://www.bookingchallenge.com/final-leaderboard-results\n\nI was lucky to win with @titericz , @cdeotte @jiweiliu and other NVIDIA colleagues. :D", "votes": 4 }, { "id": 1205655, "postDate": "2021-02-16T21:11:17.170Z", "content": "I'm very happy for you, congratulations! :))
", "rawMarkdown": "I'm very happy for you, congratulations! :))" } ] }, { "id": 1160997, "postDate": "2021-01-20T08:56:48.223Z", "content": "If I were you, i'd leave it at 1 submission without trying to improve, just for the bragging rights :D
\n(Low submission count entries are the most impressive )
Too late ;) And I'd prefer to brag about my final result if it is good than on the number of subs ;)
", "rawMarkdown": "Too late ;) And I'd prefer to brag about my final result if it is good than on the number of subs ;)", "votes": 1 } ] }, { "id": 1159960, "postDate": "2021-01-19T15:17:42.653Z", "content": "Thanks for clearing this. The scores looked to be incredibly high at first glance. However I read elsewhere that the public score corresponds to the private score well before that leak was plugged.
\nCongrats and hope you go on to win the solo gold
", "rawMarkdown": "Thanks for clearing this. The scores looked to be incredibly high at first glance. However I read elsewhere that the public score corresponds to the private score well before that leak was plugged.\n\nCongrats and hope you go on to win the solo gold", "votes": 2 }, { "id": 1159904, "postDate": "2021-01-19T14:35:01.287Z", "content": "wow thats great ! btw are you using SED method described in PANN 's Github architecture ? or doing any other method ?
", "rawMarkdown": "wow thats great ! btw are you using SED method described in PANN 's Github architecture ? or doing any other method ?", "votes": 2, "replies": [ { "id": 1159910, "postDate": "2021-01-19T14:37:39.553Z", "content": "I won't disclose more as it is the first time I am in position to get a solo prize. I hope you'll understand.
", "rawMarkdown": "I won't disclose more as it is the first time I am in position to get a solo prize. I hope you'll understand.", "votes": 17 }, { "id": 1159922, "postDate": "2021-01-19T14:45:38.397Z", "content": "yes i understand !!! anyways thanx for this post it really motivated me to do what i was doing !
", "rawMarkdown": "yes i understand !!! anyways thanx for this post it really motivated me to do what i was doing !", "votes": 1 }, { "id": 1159923, "postDate": "2021-01-19T14:45:50.700Z", "content": "Thank you for not sharing a high scoring solution's details.
", "rawMarkdown": "Thank you for not sharing a high scoring solution's details. ", "votes": 17 }, { "id": 1160487, "postDate": "2021-01-20T00:47:58.010Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1161168, "postDate": "2021-01-20T11:41:12.630Z", "content": "Awesome. Now wondering whether my most viable option is to figure out how to telepathically steal thoughts, or actually try to win this comp. It's close.
", "rawMarkdown": "Awesome. Now wondering whether my most viable option is to figure out how to telepathically steal thoughts, or actually try to win this comp. It's close.", "votes": 3 }, { "id": 1163372, "postDate": "2021-01-21T16:50:03.913Z", "content": "There is an alternative to telepathy, but it is not real time: read and learn from what top teams share after competition ends. I have been doing it since I joined Kaggle, and this is extremely valuable.
\nIt is never too late, one can look at past competitions and study top solutions. Here, it may make sense to study top solutions to the Cornell bird song competition given the topic was similar.
", "rawMarkdown": "There is an alternative to telepathy, but it is not real time: read and learn from what top teams share after competition ends. I have been doing it since I joined Kaggle, and this is extremely valuable.\n\nIt is never too late, one can look at past competitions and study top solutions. Here, it may make sense to study top solutions to the Cornell bird song competition given the topic was similar.", "votes": 7 } ] }, { "id": 1161935, "postDate": "2021-01-20T20:45:54.430Z", "content": "Great result and thank you for not sharing, I hope we can see your solution when the competition is over. 👍
", "rawMarkdown": "Great result and thank you for not sharing, I hope we can see your solution when the competition is over. 👍", "replies": [ { "id": 1161940, "postDate": "2021-01-20T20:48:43.870Z", "content": "I will share after competition end indeed.
", "rawMarkdown": "I will share after competition end indeed.", "votes": 2 } ] } ], "comments": [ { "id": 1160012, "author_name": "Theo Viel", "author_url": "", "post_date": "2021-01-19T15:58:48.443000", "content": "The fact that you were about to reach 0.931 in one submission without using any leak is already a lot of information disclosed ;)
", "votes": 9, "replies": [ { "id": 1160169, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-19T17:58:24.980000", "content": "I hope I won't regret sharing it ;)
", "votes": 8, "replies": [] } ] }, { "id": 1164109, "author_name": "Athar Sayed", "author_url": "", "post_date": "2021-01-22T06:51:36.030000", "content": "@cpmpml , I remember watching one of your talks \"Beyond Feature Engineering and HPO\" , in that you mention one user who win comp with just 1 submission may be you might be able to replicate it this time 😄😄.
", "votes": 3, "replies": [ { "id": 1164295, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-22T09:26:01.303000", "content": "Indeed, the user was @speculation and he won expedia competition with one sub indeed: https://www.kaggle.com/c/expedia-hotel-recommendations/leaderboard
\nThis is much much better than me having one sub that ranked 3rd in the middle of the competition.
", "votes": 4, "replies": [] } ] }, { "id": 1160016, "author_name": "Ahmet Erdem", "author_url": "", "post_date": "2021-01-19T16:02:15.820000", "content": "Did you expect this score or you got better/worse than you expected?
", "votes": 3, "replies": [ { "id": 1160134, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-19T17:27:51.243000", "content": "It could have been better, i.e. my CV is higher ;)
\nBut I am happy with what it is, it confirms that my baseline is sound.
", "votes": 5, "replies": [] }, { "id": 1183346, "author_name": "tugstugi", "author_url": "", "post_date": "2021-02-02T23:04:53.737000", "content": "It sounds like you are using really short segments?
", "votes": 0, "replies": [] }, { "id": 1183354, "author_name": "CPMP", "author_url": "", "post_date": "2021-02-02T23:21:16.063000", "content": "What makes you think that? Anyway, I'll try to not share what I do before competition end. It seems that the little I shared helped a lot of people already ;)
", "votes": 1, "replies": [] }, { "id": 1183360, "author_name": "tugstugi", "author_url": "", "post_date": "2021-02-02T23:30:27.707000", "content": "I only get CV +0.92 if I use shorter segments :)
", "votes": 1, "replies": [] } ] }, { "id": 1163306, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-01-21T16:28:33.603000", "content": "May I ask what a leak here means?
\nAnyways, thanks for the encouragement!
", "votes": 1, "replies": [ { "id": 1163361, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-21T16:47:24.493000", "content": "People don't always agree on what a leak is, but here is how Kaggle defines it: https://www.kaggle.com/docs/competitions#leakage
", "votes": 1, "replies": [] }, { "id": 1163438, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-01-21T17:12:42.050000", "content": "Oh I see, thank you!
", "votes": 0, "replies": [] } ] }, { "id": 1160822, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2021-01-20T06:42:54.927000", "content": "Very interesting, thanks for sharing. I might consider checking this competition then. :D
", "votes": 1, "replies": [] }, { "id": 1160267, "author_name": "Sinan Calisir", "author_url": "", "post_date": "2021-01-19T19:09:35.167000", "content": "This competition taught me a lot so far and now I can't wait to see the winning solutions. I can't even imagine what will happen then :)
", "votes": 1, "replies": [] }, { "id": 1160265, "author_name": "Sinan Calisir", "author_url": "", "post_date": "2021-01-19T19:07:09.190000", "content": "Just woow 🔥
\nOnly 1 fire because of the total submission count :)
", "votes": 1, "replies": [] }, { "id": 1160210, "author_name": "Samrath Chauhan", "author_url": "", "post_date": "2021-01-19T18:30:20.310000", "content": "Without no leak? Kudos to you👍
", "votes": 1, "replies": [] }, { "id": 1164308, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-01-22T09:37:36.117000", "content": "I feel like the top scores have gone wild since your post :D
", "votes": 2, "replies": [ { "id": 1168470, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-25T01:00:21.047000", "content": "Wait till I move to larger models than effcientnet B0 ;)
", "votes": 4, "replies": [] }, { "id": 1173871, "author_name": "MaChaogong", "author_url": "", "post_date": "2021-01-28T06:20:10.960000", "content": "I'm still waiting.
", "votes": 3, "replies": [] }, { "id": 1173929, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-01-28T07:18:09.367000", "content": "Yeah, smaller models seem to work better.
", "votes": 0, "replies": [] }, { "id": 1176061, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-29T13:33:46.227000", "content": "\n\nI'm still waiting.
\n
I spent 5 days on another (non kaggle) competition, I resumed here yesterday. I hope to get something better within next few days.
", "votes": 3, "replies": [] }, { "id": 1184312, "author_name": "CPMP", "author_url": "", "post_date": "2021-02-03T13:37:27.913000", "content": "For those interested, it was this competition https://www.bookingchallenge.com/final-leaderboard-results
\nI was lucky to win with @titericz , @cdeotte @jiweiliu and other NVIDIA colleagues. :D
", "votes": 4, "replies": [] }, { "id": 1205655, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-02-16T21:11:17.170000", "content": "I'm very happy for you, congratulations! :))
", "votes": 0, "replies": [] } ] }, { "id": 1160997, "author_name": "Dan Ofer", "author_url": "", "post_date": "2021-01-20T08:56:48.223000", "content": "If I were you, i'd leave it at 1 submission without trying to improve, just for the bragging rights :D
\n(Low submission count entries are the most impressive )
Too late ;) And I'd prefer to brag about my final result if it is good than on the number of subs ;)
", "votes": 1, "replies": [] } ] }, { "id": 1159960, "author_name": "Allohvk", "author_url": "", "post_date": "2021-01-19T15:17:42.653000", "content": "Thanks for clearing this. The scores looked to be incredibly high at first glance. However I read elsewhere that the public score corresponds to the private score well before that leak was plugged.
\nCongrats and hope you go on to win the solo gold
", "votes": 2, "replies": [] }, { "id": 1159904, "author_name": "Shubham Thapa", "author_url": "", "post_date": "2021-01-19T14:35:01.287000", "content": "wow thats great ! btw are you using SED method described in PANN 's Github architecture ? or doing any other method ?
", "votes": 2, "replies": [ { "id": 1159910, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-19T14:37:39.553000", "content": "I won't disclose more as it is the first time I am in position to get a solo prize. I hope you'll understand.
", "votes": 17, "replies": [] }, { "id": 1159922, "author_name": "Shubham Thapa", "author_url": "", "post_date": "2021-01-19T14:45:38.397000", "content": "yes i understand !!! anyways thanx for this post it really motivated me to do what i was doing !
", "votes": 1, "replies": [] }, { "id": 1159923, "author_name": "Peter", "author_url": "", "post_date": "2021-01-19T14:45:50.700000", "content": "Thank you for not sharing a high scoring solution's details.
", "votes": 17, "replies": [] }, { "id": 1160487, "author_name": "", "author_url": "", "post_date": "2021-01-20T00:47:58.010000", "content": "", "votes": 0, "replies": [] }, { "id": 1161168, "author_name": "Alexander Soare", "author_url": "", "post_date": "2021-01-20T11:41:12.630000", "content": "Awesome. Now wondering whether my most viable option is to figure out how to telepathically steal thoughts, or actually try to win this comp. It's close.
", "votes": 3, "replies": [] }, { "id": 1163372, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-21T16:50:03.913000", "content": "There is an alternative to telepathy, but it is not real time: read and learn from what top teams share after competition ends. I have been doing it since I joined Kaggle, and this is extremely valuable.
\nIt is never too late, one can look at past competitions and study top solutions. Here, it may make sense to study top solutions to the Cornell bird song competition given the topic was similar.
", "votes": 7, "replies": [] } ] }, { "id": 1161935, "author_name": "", "author_url": "", "post_date": "2021-01-20T20:45:54.430000", "content": "Great result and thank you for not sharing, I hope we can see your solution when the competition is over. 👍
", "votes": 0, "replies": [ { "id": 1161940, "author_name": "CPMP", "author_url": "", "post_date": "2021-01-20T20:48:43.870000", "content": "I will share after competition end indeed.
", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "1159884": "As many, I wondered if top score was not due to some leak. I now don't think there is a leak because my first sub at LB 0.931 is just a 5 folds single model without any leak nor external data. \n\nI am sharing not to brag, but rather to give hope to all those who could be discouraged by the incredible top score.", "1160012": "The fact that you were about to reach 0.931 in one submission without using any leak is already a lot of information disclosed ;)", "1164109": "@cpmpml , I remember watching one of your talks \"Beyond Feature Engineering and HPO\" , in that you mention one user who win comp with just 1 submission may be you might be able to replicate it this time 😄😄.", "1160016": "Did you expect this score or you got better/worse than you expected?", "1163306": "May I ask what a leak here means?\n\nAnyways, thanks for the encouragement!", "1160822": "Very interesting, thanks for sharing. I might consider checking this competition then. :D", "1160267": "This competition taught me a lot so far and now I can't wait to see the winning solutions. I can't even imagine what will happen then :)", "1160265": "Just woow 🔥\n\nOnly 1 fire because of the total submission count :)", "1160210": "Without no leak? Kudos to you👍", "1164308": "I feel like the top scores have gone wild since your post :D", "1160997": "If I were you, i'd leave it at 1 submission without trying to improve, just for the bragging rights :D \n(Low submission count entries are the most impressive )", "1159960": "Thanks for clearing this. The scores looked to be incredibly high at first glance. However I read elsewhere that the public score corresponds to the private score well before that leak was plugged.\n\nCongrats and hope you go on to win the solo gold", "1159904": "wow thats great ! btw are you using SED method described in PANN 's Github architecture ? or doing any other method ?", "1161935": "Great result and thank you for not sharing, I hope we can see your solution when the competition is over. 👍" } }