Here in ODS.ai we respect Kaggle community.
\n\nMy talk at Datafest is about this competition and my solutions.
\n\nBecause organizers extended the competition deadline, we share live video of Datafest in Odessa in order to comply with Kaggle rules.
\n\nWatch us live!\nhttps://youtu.be/gWYVSH0kZEk [russian]
", "messages": [ { "id": 620246, "postDate": "2019-09-07T07:49:55.853Z", "content": "Here in ODS.ai we respect Kaggle community.
\n\nMy talk at Datafest is about this competition and my solutions.
\n\nBecause organizers extended the competition deadline, we share live video of Datafest in Odessa in order to comply with Kaggle rules.
\n\nWatch us live!\nhttps://youtu.be/gWYVSH0kZEk [russian]
", "rawMarkdown": "Here in ODS.ai we respect Kaggle community.\n\nMy talk at Datafest is about this competition and my solutions. \n\nBecause organizers extended the competition deadline, we share live video of Datafest in Odessa in order to comply with Kaggle rules.\n\nWatch us live!\nhttps://youtu.be/gWYVSH0kZEk [russian]", "votes": 52 }, { "id": 620279, "postDate": "2019-09-07T08:41:34.947Z", "content": "Here are slides, too
", "rawMarkdown": "Here are slides, too", "votes": 12 }, { "id": 620405, "postDate": "2019-09-07T13:09:42.143Z", "content": "Great slides! One question Why you were exchanging head of NN after pretraining on 2015 dataest?
", "rawMarkdown": "Great slides! One question Why you were exchanging head of NN after pretraining on 2015 dataest? ", "votes": 3, "replies": [ { "id": 620677, "postDate": "2019-09-07T20:23:20.920Z", "content": "Thank you!\nWe observed, that new head scored consistently better CV than finetuning head from 2015 data, so we decided to make it anew every time
", "rawMarkdown": "Thank you!\nWe observed, that new head scored consistently better CV than finetuning head from 2015 data, so we decided to make it anew every time", "votes": 1 }, { "id": 620686, "postDate": "2019-09-07T20:51:18.723Z", "content": "Using my pipeline the new head gives slight worse CV but better LB than finetuning head from 2015 data. Perhaps it is related to some other features.
", "rawMarkdown": "Using my pipeline the new head gives slight worse CV but better LB than finetuning head from 2015 data. Perhaps it is related to some other features. ", "votes": 1 }, { "id": 620688, "postDate": "2019-09-07T21:06:57.203Z", "content": "Interesting, should try that. It sounds weird to me though.
", "rawMarkdown": "Interesting, should try that. It sounds weird to me though." } ] }, { "id": 620311, "postDate": "2019-09-07T09:47:23.127Z", "content": "I just glanced through, and it seems that most of your experiments were already mentioned in discussions or public kernels.
", "rawMarkdown": "I just glanced through, and it seems that most of your experiments were already mentioned in discussions or public kernels.", "votes": 3, "replies": [ { "id": 620317, "postDate": "2019-09-07T09:56:05.057Z", "content": "Isn't this true for most participants' solutions?
", "rawMarkdown": "Isn't this true for most participants' solutions?" }, { "id": 620326, "postDate": "2019-09-07T10:19:22.093Z", "content": "Yes, of course. That's also why the gold is out of reach for most participants.
", "rawMarkdown": "Yes, of course. That's also why the gold is out of reach for most participants." } ] }, { "id": 620376, "postDate": "2019-09-07T12:14:21.180Z", "content": "That's a great talk! Thanks for sharing and good luck for tomorrow!
", "rawMarkdown": "That's a great talk! Thanks for sharing and good luck for tomorrow!", "votes": 1 }, { "id": 620248, "postDate": "2019-09-07T07:52:02.723Z", "content": "If you could translate everything to English (slides are in English) and maybe make the post not seconds before starting the talk it would be better... Actually not doing the talk at all before competition ends would be the proper thing to do, I think everyone at the event would understand.
", "rawMarkdown": "If you could translate everything to English (slides are in English) and maybe make the post not seconds before starting the talk it would be better... Actually not doing the talk at all before competition ends would be the proper thing to do, I think everyone at the event would understand.", "votes": -4, "replies": [ { "id": 620294, "postDate": "2019-09-07T09:02:58.860Z", "content": "I suppose this talk was planned long ago and deadline extension was a surprise. It was difficult to prepare a different talk instead of this one, so they left it as it is.
", "rawMarkdown": "I suppose this talk was planned long ago and deadline extension was a surprise. It was difficult to prepare a different talk instead of this one, so they left it as it is.", "votes": 11 }, { "id": 620298, "postDate": "2019-09-07T09:20:36.337Z", "content": "Fair points, I can understand. At least then posting it here earlier would have been kind.
", "rawMarkdown": "Fair points, I can understand. At least then posting it here earlier would have been kind.", "votes": 1 } ] }, { "id": 620487, "postDate": "2019-09-07T15:03:17.843Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 620737, "postDate": "2019-09-07T23:33:33.947Z", "content": "Thanks for sharing
", "rawMarkdown": "Thanks for sharing \n\n", "votes": 1 }, { "id": 620518, "postDate": "2019-09-07T15:50:37.457Z", "content": "Great! Keep it up. Thanks for sharing.
", "rawMarkdown": "Great! Keep it up. Thanks for sharing.", "votes": 1 } ], "comments": [ { "id": 620279, "author_name": "Borys Tymchenko", "author_url": "", "post_date": "2019-09-07T08:41:34.947000", "content": "Here are slides, too
", "votes": 12, "replies": [] }, { "id": 620405, "author_name": "DrHB", "author_url": "", "post_date": "2019-09-07T13:09:42.143000", "content": "Great slides! One question Why you were exchanging head of NN after pretraining on 2015 dataest?
", "votes": 3, "replies": [ { "id": 620677, "author_name": "Borys Tymchenko", "author_url": "", "post_date": "2019-09-07T20:23:20.920000", "content": "Thank you!\nWe observed, that new head scored consistently better CV than finetuning head from 2015 data, so we decided to make it anew every time
", "votes": 1, "replies": [] }, { "id": 620686, "author_name": "Xuan Cao", "author_url": "", "post_date": "2019-09-07T20:51:18.723000", "content": "Using my pipeline the new head gives slight worse CV but better LB than finetuning head from 2015 data. Perhaps it is related to some other features.
", "votes": 1, "replies": [] }, { "id": 620688, "author_name": "Psi", "author_url": "", "post_date": "2019-09-07T21:06:57.203000", "content": "Interesting, should try that. It sounds weird to me though.
", "votes": 0, "replies": [] } ] }, { "id": 620311, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-07T09:47:23.127000", "content": "I just glanced through, and it seems that most of your experiments were already mentioned in discussions or public kernels.
", "votes": 3, "replies": [ { "id": 620317, "author_name": "Homoalways", "author_url": "", "post_date": "2019-09-07T09:56:05.057000", "content": "Isn't this true for most participants' solutions?
", "votes": 0, "replies": [] }, { "id": 620326, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-07T10:19:22.093000", "content": "Yes, of course. That's also why the gold is out of reach for most participants.
", "votes": 0, "replies": [] } ] }, { "id": 620376, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-07T12:14:21.180000", "content": "That's a great talk! Thanks for sharing and good luck for tomorrow!
", "votes": 1, "replies": [] }, { "id": 620248, "author_name": "Psi", "author_url": "", "post_date": "2019-09-07T07:52:02.723000", "content": "If you could translate everything to English (slides are in English) and maybe make the post not seconds before starting the talk it would be better... Actually not doing the talk at all before competition ends would be the proper thing to do, I think everyone at the event would understand.
", "votes": -4, "replies": [ { "id": 620294, "author_name": "Andrey Lukyanenko", "author_url": "", "post_date": "2019-09-07T09:02:58.860000", "content": "I suppose this talk was planned long ago and deadline extension was a surprise. It was difficult to prepare a different talk instead of this one, so they left it as it is.
", "votes": 11, "replies": [] }, { "id": 620298, "author_name": "Psi", "author_url": "", "post_date": "2019-09-07T09:20:36.337000", "content": "Fair points, I can understand. At least then posting it here earlier would have been kind.
", "votes": 1, "replies": [] } ] }, { "id": 620487, "author_name": "", "author_url": "", "post_date": "2019-09-07T15:03:17.843000", "content": "", "votes": 1, "replies": [] }, { "id": 620737, "author_name": "forwardPropagation", "author_url": "", "post_date": "2019-09-07T23:33:33.947000", "content": "Thanks for sharing
", "votes": 1, "replies": [] }, { "id": 620518, "author_name": "Rashidul H", "author_url": "", "post_date": "2019-09-07T15:50:37.457000", "content": "Great! Keep it up. Thanks for sharing.
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "620246": "Here in ODS.ai we respect Kaggle community.\n\nMy talk at Datafest is about this competition and my solutions. \n\nBecause organizers extended the competition deadline, we share live video of Datafest in Odessa in order to comply with Kaggle rules.\n\nWatch us live!\nhttps://youtu.be/gWYVSH0kZEk [russian]", "620279": "Here are slides, too", "620405": "Great slides! One question Why you were exchanging head of NN after pretraining on 2015 dataest? ", "620311": "I just glanced through, and it seems that most of your experiments were already mentioned in discussions or public kernels.", "620376": "That's a great talk! Thanks for sharing and good luck for tomorrow!", "620248": "If you could translate everything to English (slides are in English) and maybe make the post not seconds before starting the talk it would be better... Actually not doing the talk at all before competition ends would be the proper thing to do, I think everyone at the event would understand.", "620487": "", "620737": "Thanks for sharing \n\n", "620518": "Great! Keep it up. Thanks for sharing." } }