Anyone of us who has worked on this problem for a long while knows too well how hard it is to improve the LB by 0.001 ... (In my case, I can do 0.001 improvement / week — truly slow progress)
\n\nSuddenly, Pavel then joined us ... and boom!
\n\nSeeing what LB score his team can achieve ... all I can say is ... \nPavel and Evgeny, Looking forward to learn from you at the end of the competition!
\n\nAs well as many incredible teams here! Please keep up your great works!
", "messages": [ { "id": 462667, "postDate": "2019-01-28T17:20:24.513Z", "content": "I'd be more surprised if they survived shakeup with the same score. Public LB is not that informative here
", "rawMarkdown": "I'd be more surprised if they survived shakeup with the same score. Public LB is not that informative here", "votes": 12, "replies": [ { "id": 463622, "postDate": "2019-01-30T09:57:11.283Z", "content": "Hi Yury, you may be right, but even the case of overfitting LB, I still want to know how to do it ;)
", "rawMarkdown": "Hi Yury, you may be right, but even the case of overfitting LB, I still want to know how to do it ;)", "votes": 1 }, { "id": 463687, "postDate": "2019-01-30T12:32:09.340Z", "content": "In principle, there are ways to cheat, that won't generalize to Stage 2. Take a look at these 2 kernels: one and two.
", "rawMarkdown": "In principle, there are ways to cheat, that won't generalize to Stage 2. Take a look at these 2 kernels: [one][1] and [two][2].\n\n\n [1]: https://www.kaggle.com/ishitori/what-if-we-could-finetune-bert-from-apache-mxnet\n [2]: https://www.kaggle.com/paulorzp/compressing-a-binary-submission-within-a-string", "votes": 1 }, { "id": 464773, "postDate": "2019-02-01T13:34:41.583Z", "content": "Thanks for your information Yury! Nevertheless, I am not assuming that they are cheating. My assumption is that perhaps they have some secret tricks or architectures (well, hope to see the truth at the end :)
", "rawMarkdown": "Thanks for your information Yury! Nevertheless, I am not assuming that they are cheating. My assumption is that perhaps they have some secret tricks or architectures (well, hope to see the truth at the end :)" }, { "id": 466862, "postDate": "2019-02-06T05:13:58.047Z", "content": "Deleted<> I just saw it now 0.786. Secret trick indeed. Yes, they are genuis.
", "rawMarkdown": "Deleted<" } ] }, { "id": 462623, "postDate": "2019-01-28T15:55:52.923Z", "content": "Anyone of us who has worked on this problem for a long while knows too well how hard it is to improve the LB by 0.001 ... (In my case, I can do 0.001 improvement / week — truly slow progress)
\n\nSuddenly, Pavel then joined us ... and boom!
\n\nSeeing what LB score his team can achieve ... all I can say is ... \nPavel and Evgeny, Looking forward to learn from you at the end of the competition!
\n\nAs well as many incredible teams here! Please keep up your great works!
", "rawMarkdown": "Anyone of us who has worked on this problem for a long while knows too well how hard it is to improve the LB by 0.001 ... (In my case, I can do 0.001 improvement / week — truly slow progress)\n\nSuddenly, Pavel then joined us ... and boom!\n\nSeeing what LB score his team can achieve ... all I can say is ... \nPavel and Evgeny, Looking forward to learn from you at the end of the competition! \n\nAs well as many incredible teams here! Please keep up your great works!", "votes": 11 }, { "id": 462634, "postDate": "2019-01-28T16:15:06.583Z", "content": "woke up this morning to see that and felt a mixture of being amazed / laughing and ready to throw my computer out the window :D
", "rawMarkdown": "woke up this morning to see that and felt a mixture of being amazed / laughing and ready to throw my computer out the window :D", "votes": 5, "replies": [ { "id": 462641, "postDate": "2019-01-28T16:28:58.757Z", "content": "Ha ha ... My desktop is indeed quite old, and maybe I have an excuse for a new one :p
", "rawMarkdown": "Ha ha ... My desktop is indeed quite old, and maybe I have an excuse for a new one :p", "votes": 3 }, { "id": 462947, "postDate": "2019-01-29T06:29:01.897Z", "content": "This competition is surely a roller coster ride for me. The same kernel gives lot of variation on public LB. Right now we are at our best CV of 0.701 and Public LB at 0.699
\n\nA stable validation in place. Trusting in CV till the end. Let's see how it will wind up :)
", "rawMarkdown": "This competition is surely a roller coster ride for me. The same kernel gives lot of variation on public LB. Right now we are at our best CV of 0.701 and Public LB at 0.699\n\nA stable validation in place. Trusting in CV till the end. Let's see how it will wind up :) ", "votes": 5 }, { "id": 462968, "postDate": "2019-01-29T06:52:42.827Z", "content": "Even though it still remains around one week for performance improvement by myself, I can’t wait to learn great tricks from many peers here ... Wish the best of luck for us!
", "rawMarkdown": "Even though it still remains around one week for performance improvement by myself, I can’t wait to learn great tricks from many peers here ... Wish the best of luck for us!", "votes": 1 }, { "id": 462976, "postDate": "2019-01-29T07:13:13.657Z", "content": "How do you calculate your cv score? Is it the score on a holdout set?, or is your cv strategy more like https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77949 ?
", "rawMarkdown": "How do you calculate your cv score? Is it the score on a holdout set?, or is your cv strategy more like https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77949 ?", "votes": 1 } ] }, { "id": 466402, "postDate": "2019-02-05T10:29:20.757Z", "content": "now , it is 0.786
", "rawMarkdown": "now , it is 0.786", "votes": 3, "replies": [ { "id": 466417, "postDate": "2019-02-05T10:53:47.277Z", "content": "What a super trick!
", "rawMarkdown": "What a super trick!" }, { "id": 466485, "postDate": "2019-02-05T13:21:29.130Z", "content": "I... I give up...
", "rawMarkdown": "I... I give up...", "votes": 1 }, { "id": 466820, "postDate": "2019-02-06T01:47:13.567Z", "content": "Hi Max, hope you have done well! For my part, I am not so sure ;)
", "rawMarkdown": "Hi Max, hope you have done well! For my part, I am not so sure ;) ", "votes": 1 }, { "id": 467037, "postDate": "2019-02-06T11:18:42.833Z", "content": "Neither am I! This competition is too stressful >.< The noise is so big that there's absolutely no telling if all that work has been worth it... Let's hope for the best!
", "rawMarkdown": "Neither am I! This competition is too stressful >.< The noise is so big that there's absolutely no telling if all that work has been worth it... Let's hope for the best!", "votes": 1 } ] }, { "id": 462632, "postDate": "2019-01-28T16:11:35.813Z", "content": "Yeah, their score is really amazing but I'm more curious about The zoo team's solution. Because they improved their score steadily.
", "rawMarkdown": "Yeah, their score is really amazing but I'm more curious about The zoo team's solution. Because they improved their score steadily.", "votes": 3, "replies": [ { "id": 462639, "postDate": "2019-01-28T16:27:36.483Z", "content": "I am sorry not to mention them first. Just got too shocked. The Zoo team is indeed great and they are here from the beginning (they have worked so hard for 3 months). I am truly respect them as well as many people who are with us during these long marathon (or rather hackathon).
\n\nLooking forward to learn from them too!
", "rawMarkdown": "I am sorry not to mention them first. Just got too shocked. The Zoo team is indeed great and they are here from the beginning (they have worked so hard for 3 months). I am truly respect them as well as many people who are with us during these long marathon (or rather hackathon). \n\nLooking forward to learn from them too!", "votes": 9 } ] }, { "id": 465325, "postDate": "2019-02-02T20:31:39.007Z", "content": "Now they have the LB score of 72.9%...\nTotally unbelievable!
", "rawMarkdown": "Now they have the LB score of 72.9%...\nTotally unbelievable!", "votes": 2 }, { "id": 466401, "postDate": "2019-02-05T10:27:06.457Z", "content": "And now it's... amazing
", "rawMarkdown": "And now it's... amazing", "votes": 1 }, { "id": 465377, "postDate": "2019-02-03T00:48:39.050Z", "content": "Now，the score is 0.729
", "rawMarkdown": "Now，the score is 0.729", "votes": 1, "replies": [ { "id": 465396, "postDate": "2019-02-03T02:23:04.463Z", "content": "That's an amazing degree of improvement this late in the competition. Let's see how they do on the private set...
", "rawMarkdown": "That's an amazing degree of improvement this late in the competition. Let's see how they do on the private set...", "votes": 1 } ] }, { "id": 463650, "postDate": "2019-01-30T11:21:27.147Z", "content": "Truly impressive. While initially frustrating, this competition pushed me hard and I am grateful for it.
", "rawMarkdown": "Truly impressive. While initially frustrating, this competition pushed me hard and I am grateful for it.", "votes": 2 }, { "id": 466421, "postDate": "2019-02-05T11:01:12.473Z", "content": "truly amazing!
", "rawMarkdown": "truly amazing!" }, { "id": 466407, "postDate": "2019-02-05T10:41:52.470Z", "content": "We need Discussion (Flood) Tab ...
", "rawMarkdown": "We need Discussion (Flood) Tab ... ", "replies": [ { "id": 466413, "postDate": "2019-02-05T10:50:22.470Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 462631, "postDate": "2019-01-28T16:09:42.807Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 462667, "author_name": "Yury Kashnitsky", "author_url": "", "post_date": "2019-01-28T17:20:24.513000", "content": "I'd be more surprised if they survived shakeup with the same score. Public LB is not that informative here
", "votes": 12, "replies": [ { "id": 463622, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-01-30T09:57:11.283000", "content": "Hi Yury, you may be right, but even the case of overfitting LB, I still want to know how to do it ;)
", "votes": 1, "replies": [] }, { "id": 463687, "author_name": "Yury Kashnitsky", "author_url": "", "post_date": "2019-01-30T12:32:09.340000", "content": "In principle, there are ways to cheat, that won't generalize to Stage 2. Take a look at these 2 kernels: one and two.
", "votes": 1, "replies": [] }, { "id": 464773, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-02-01T13:34:41.583000", "content": "Thanks for your information Yury! Nevertheless, I am not assuming that they are cheating. My assumption is that perhaps they have some secret tricks or architectures (well, hope to see the truth at the end :)
", "votes": 0, "replies": [] }, { "id": 466862, "author_name": "Kepler456b", "author_url": "", "post_date": "2019-02-06T05:13:58.047000", "content": "Deleted<> I just saw it now 0.786. Secret trick indeed. Yes, they are genuis.
", "votes": 0, "replies": [] } ] }, { "id": 462634, "author_name": "eee", "author_url": "", "post_date": "2019-01-28T16:15:06.583000", "content": "woke up this morning to see that and felt a mixture of being amazed / laughing and ready to throw my computer out the window :D
", "votes": 5, "replies": [ { "id": 462641, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-01-28T16:28:58.757000", "content": "Ha ha ... My desktop is indeed quite old, and maybe I have an excuse for a new one :p
", "votes": 3, "replies": [] }, { "id": 462947, "author_name": "Shahebaz Mohammad", "author_url": "", "post_date": "2019-01-29T06:29:01.897000", "content": "This competition is surely a roller coster ride for me. The same kernel gives lot of variation on public LB. Right now we are at our best CV of 0.701 and Public LB at 0.699
\n\nA stable validation in place. Trusting in CV till the end. Let's see how it will wind up :)
", "votes": 5, "replies": [] }, { "id": 462968, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-01-29T06:52:42.827000", "content": "Even though it still remains around one week for performance improvement by myself, I can’t wait to learn great tricks from many peers here ... Wish the best of luck for us!
", "votes": 1, "replies": [] }, { "id": 462976, "author_name": "bilal2vec", "author_url": "", "post_date": "2019-01-29T07:13:13.657000", "content": "How do you calculate your cv score? Is it the score on a holdout set?, or is your cv strategy more like https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77949 ?
", "votes": 1, "replies": [] } ] }, { "id": 466402, "author_name": "LiuDong", "author_url": "", "post_date": "2019-02-05T10:29:20.757000", "content": "now , it is 0.786
", "votes": 3, "replies": [ { "id": 466417, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-02-05T10:53:47.277000", "content": "What a super trick!
", "votes": 0, "replies": [] }, { "id": 466485, "author_name": "Max Schumacher", "author_url": "", "post_date": "2019-02-05T13:21:29.130000", "content": "I... I give up...
", "votes": 1, "replies": [] }, { "id": 466820, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-02-06T01:47:13.567000", "content": "Hi Max, hope you have done well! For my part, I am not so sure ;)
", "votes": 1, "replies": [] }, { "id": 467037, "author_name": "Max Schumacher", "author_url": "", "post_date": "2019-02-06T11:18:42.833000", "content": "Neither am I! This competition is too stressful >.< The noise is so big that there's absolutely no telling if all that work has been worth it... Let's hope for the best!
", "votes": 1, "replies": [] } ] }, { "id": 462632, "author_name": "Murat Korkmaz", "author_url": "", "post_date": "2019-01-28T16:11:35.813000", "content": "Yeah, their score is really amazing but I'm more curious about The zoo team's solution. Because they improved their score steadily.
", "votes": 3, "replies": [ { "id": 462639, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-01-28T16:27:36.483000", "content": "I am sorry not to mention them first. Just got too shocked. The Zoo team is indeed great and they are here from the beginning (they have worked so hard for 3 months). I am truly respect them as well as many people who are with us during these long marathon (or rather hackathon).
\n\nLooking forward to learn from them too!
", "votes": 9, "replies": [] } ] }, { "id": 465325, "author_name": "yufuin", "author_url": "", "post_date": "2019-02-02T20:31:39.007000", "content": "Now they have the LB score of 72.9%...\nTotally unbelievable!
", "votes": 2, "replies": [] }, { "id": 466401, "author_name": "cfiken", "author_url": "", "post_date": "2019-02-05T10:27:06.457000", "content": "And now it's... amazing
", "votes": 1, "replies": [] }, { "id": 465377, "author_name": "Junxiong Liu", "author_url": "", "post_date": "2019-02-03T00:48:39.050000", "content": "Now，the score is 0.729
", "votes": 1, "replies": [ { "id": 465396, "author_name": "derplearning", "author_url": "", "post_date": "2019-02-03T02:23:04.463000", "content": "That's an amazing degree of improvement this late in the competition. Let's see how they do on the private set...
", "votes": 1, "replies": [] } ] }, { "id": 463650, "author_name": "Bjenk Ellefsen", "author_url": "", "post_date": "2019-01-30T11:21:27.147000", "content": "Truly impressive. While initially frustrating, this competition pushed me hard and I am grateful for it.
", "votes": 2, "replies": [] }, { "id": 466421, "author_name": "Soonhwan Kwon", "author_url": "", "post_date": "2019-02-05T11:01:12.473000", "content": "truly amazing!
", "votes": 0, "replies": [] }, { "id": 466407, "author_name": "Oleg Yaroshevskiy", "author_url": "", "post_date": "2019-02-05T10:41:52.470000", "content": "We need Discussion (Flood) Tab ...
", "votes": 0, "replies": [ { "id": 466413, "author_name": "", "author_url": "", "post_date": "2019-02-05T10:50:22.470000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 462631, "author_name": "", "author_url": "", "post_date": "2019-01-28T16:09:42.807000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "462667": "I'd be more surprised if they survived shakeup with the same score. Public LB is not that informative here", "462623": "Anyone of us who has worked on this problem for a long while knows too well how hard it is to improve the LB by 0.001 ... (In my case, I can do 0.001 improvement / week — truly slow progress)\n\nSuddenly, Pavel then joined us ... and boom!\n\nSeeing what LB score his team can achieve ... all I can say is ... \nPavel and Evgeny, Looking forward to learn from you at the end of the competition! \n\nAs well as many incredible teams here! Please keep up your great works!", "462634": "woke up this morning to see that and felt a mixture of being amazed / laughing and ready to throw my computer out the window :D", "466402": "now , it is 0.786", "462632": "Yeah, their score is really amazing but I'm more curious about The zoo team's solution. Because they improved their score steadily.", "465325": "Now they have the LB score of 72.9%...\nTotally unbelievable!", "466401": "And now it's... amazing", "465377": "Now，the score is 0.729", "463650": "Truly impressive. While initially frustrating, this competition pushed me hard and I am grateful for it.", "466421": "truly amazing!", "466407": "We need Discussion (Flood) Tab ... ", "462631": "" } }