Follow DrHB's suggestion, use B3 net work, more on clean the 2019 data and select on 2015 data, and preprocess. one model get 0.807 with cv 0.92. If your approach is different and have similar score, maybe we could ensemble our models to get high score and generalize our approaches to avoid shakeup in private data set.
", "messages": [ { "id": 596063, "postDate": "2019-08-10T05:31:02.397Z", "content": "Follow DrHB's suggestion, use B3 net work, more on clean the 2019 data and select on 2015 data, and preprocess. one model get 0.807 with cv 0.92. If your approach is different and have similar score, maybe we could ensemble our models to get high score and generalize our approaches to avoid shakeup in private data set.
", "rawMarkdown": "Follow DrHB's suggestion, use B3 net work, more on clean the 2019 data and select on 2015 data, and preprocess. one model get 0.807 with cv 0.92. If your approach is different and have similar score, maybe we could ensemble our models to get high score and generalize our approaches to avoid shakeup in private data set.", "votes": 2 }, { "id": 607154, "postDate": "2019-08-24T17:12:15.260Z", "content": "I have one simple model based on 2019 data (LB0.818) with the new GPU restriction it will be too hard so if someone wants to team let me know
", "rawMarkdown": "I have one simple model based on 2019 data (LB0.818) with the new GPU restriction it will be too hard so if someone wants to team let me know" }, { "id": 606403, "postDate": "2019-08-23T14:33:42.260Z", "content": "Why there is this restriction?
\n\nThe combined submission total (331) exceeds the maximum allowed (285). Unfortunately this restriction will likely apply up to the competition's team merge deadline.
", "rawMarkdown": "Why there is this restriction?\n\nThe combined submission total (331) exceeds the maximum allowed (285). Unfortunately this restriction will likely apply up to the competition's team merge deadline." }, { "id": 605950, "postDate": "2019-08-23T02:01:59.877Z", "content": "I'd like to team up. Best single model is 0.807 DenseNet with ordinal regression.
", "rawMarkdown": "I'd like to team up. Best single model is 0.807 DenseNet with ordinal regression. ", "replies": [ { "id": 606407, "postDate": "2019-08-23T14:38:38.527Z", "content": "@lextoumbourou I'd like to form a team too
", "rawMarkdown": "@lextoumbourou I'd like to form a team too" }, { "id": 606420, "postDate": "2019-08-23T14:54:58.280Z", "content": "I have a single efficientnet model with 0.794 LB. Would u like to team up ?
", "rawMarkdown": "I have a single efficientnet model with 0.794 LB. Would u like to team up ?" } ] }, { "id": 597063, "postDate": "2019-08-11T18:41:32.217Z", "content": "I'd also like to team up, single (efficientnet) model 0.79LB, 0.9CV on 2019 data only, no cleaning
", "rawMarkdown": "I'd also like to team up, single (efficientnet) model 0.79LB, 0.9CV on 2019 data only, no cleaning", "replies": [ { "id": 597514, "postDate": "2019-08-12T12:52:37.767Z", "content": "I have a single efficientnet model with 0.794 LB. How did you reach your current LB, ensembles? If yes, how many models?
", "rawMarkdown": "I have a single efficientnet model with 0.794 LB. How did you reach your current LB, ensembles? If yes, how many models? " }, { "id": 597620, "postDate": "2019-08-12T15:11:59.207Z", "content": "my single (efficientnet) model fluctuates between 0.79~0.799lb, a simple 3-fold cv averaging gives my current lb
", "rawMarkdown": "my single (efficientnet) model fluctuates between 0.79~0.799lb, a simple 3-fold cv averaging gives my current lb" } ] }, { "id": 596673, "postDate": "2019-08-11T05:12:41.553Z", "content": "What do you mean by clean the 2019 data?
", "rawMarkdown": "What do you mean by clean the 2019 data? ", "replies": [ { "id": 596774, "postDate": "2019-08-11T09:20:07.493Z", "content": "There are many duplicated images in 2019 data.
", "rawMarkdown": "There are many duplicated images in 2019 data.", "votes": 1 }, { "id": 596778, "postDate": "2019-08-11T09:28:24.263Z", "content": "Oh thanks . I'll look into it
", "rawMarkdown": "Oh thanks . I'll look into it" } ] }, { "id": 596650, "postDate": "2019-08-11T03:42:06.677Z", "content": "Hi I want to team up. I got higher CV score, but lower LB score. My highest single model is 0.799 with CV 0.90. Earger to exchange with ideas to decrease the CV gap and ensemble methods.
\n\nThe best CV score is based on resnext with heavy augmentation.
", "rawMarkdown": "Hi I want to team up. I got higher CV score, but lower LB score. My highest single model is 0.799 with CV 0.90. Earger to exchange with ideas to decrease the CV gap and ensemble methods.\n\nThe best CV score is based on resnext with heavy augmentation.", "replies": [ { "id": 596775, "postDate": "2019-08-11T09:21:25.797Z", "content": "Cool, let's do it
", "rawMarkdown": "Cool, let's do it" }, { "id": 597013, "postDate": "2019-08-11T16:50:30.577Z", "rawMarkdown": "", "isDeleted": true }, { "id": 597027, "postDate": "2019-08-11T17:08:51.853Z", "content": "I have requested to merge. Fighting!
", "rawMarkdown": "I have requested to merge. Fighting!" }, { "id": 597221, "postDate": "2019-08-12T02:56:21.510Z", "content": "what's your linkedin or QQ?
", "rawMarkdown": "what's your linkedin or QQ?" }, { "id": 597225, "postDate": "2019-08-12T03:09:18.437Z", "rawMarkdown": "", "isDeleted": true }, { "id": 597504, "postDate": "2019-08-12T12:32:23.557Z", "content": "@abnerzhang @xiaojiu1414 I'm also interested joining the team, please let me know what you think?
", "rawMarkdown": "@abnerzhang @xiaojiu1414 I'm also interested joining the team, please let me know what you think?" } ] }, { "id": 606443, "postDate": "2019-08-23T15:27:56.160Z", "rawMarkdown": "", "isDeleted": true }, { "id": 605910, "postDate": "2019-08-23T00:22:29.883Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 607154, "author_name": "jayjay", "author_url": "", "post_date": "2019-08-24T17:12:15.260000", "content": "I have one simple model based on 2019 data (LB0.818) with the new GPU restriction it will be too hard so if someone wants to team let me know
", "votes": 0, "replies": [] }, { "id": 606403, "author_name": "abnerzhang", "author_url": "", "post_date": "2019-08-23T14:33:42.260000", "content": "Why there is this restriction?
\n\nThe combined submission total (331) exceeds the maximum allowed (285). Unfortunately this restriction will likely apply up to the competition's team merge deadline.
", "votes": 0, "replies": [] }, { "id": 605950, "author_name": "Lex Toumbourou", "author_url": "", "post_date": "2019-08-23T02:01:59.877000", "content": "I'd like to team up. Best single model is 0.807 DenseNet with ordinal regression.
", "votes": 0, "replies": [ { "id": 606407, "author_name": "samshipengs", "author_url": "", "post_date": "2019-08-23T14:38:38.527000", "content": "@lextoumbourou I'd like to form a team too
", "votes": 0, "replies": [] }, { "id": 606420, "author_name": "Viraj Bagal", "author_url": "", "post_date": "2019-08-23T14:54:58.280000", "content": "I have a single efficientnet model with 0.794 LB. Would u like to team up ?
", "votes": 0, "replies": [] } ] }, { "id": 597063, "author_name": "samshipengs", "author_url": "", "post_date": "2019-08-11T18:41:32.217000", "content": "I'd also like to team up, single (efficientnet) model 0.79LB, 0.9CV on 2019 data only, no cleaning
", "votes": 0, "replies": [ { "id": 597514, "author_name": "Viraj Bagal", "author_url": "", "post_date": "2019-08-12T12:52:37.767000", "content": "I have a single efficientnet model with 0.794 LB. How did you reach your current LB, ensembles? If yes, how many models?
", "votes": 0, "replies": [] }, { "id": 597620, "author_name": "samshipengs", "author_url": "", "post_date": "2019-08-12T15:11:59.207000", "content": "my single (efficientnet) model fluctuates between 0.79~0.799lb, a simple 3-fold cv averaging gives my current lb
", "votes": 0, "replies": [] } ] }, { "id": 596673, "author_name": "Viraj Bagal", "author_url": "", "post_date": "2019-08-11T05:12:41.553000", "content": "What do you mean by clean the 2019 data?
", "votes": 0, "replies": [ { "id": 596774, "author_name": "abnerzhang", "author_url": "", "post_date": "2019-08-11T09:20:07.493000", "content": "There are many duplicated images in 2019 data.
", "votes": 1, "replies": [] }, { "id": 596778, "author_name": "Viraj Bagal", "author_url": "", "post_date": "2019-08-11T09:28:24.263000", "content": "Oh thanks . I'll look into it
", "votes": 0, "replies": [] } ] }, { "id": 596650, "author_name": "DeepInvolution", "author_url": "", "post_date": "2019-08-11T03:42:06.677000", "content": "Hi I want to team up. I got higher CV score, but lower LB score. My highest single model is 0.799 with CV 0.90. Earger to exchange with ideas to decrease the CV gap and ensemble methods.
\n\nThe best CV score is based on resnext with heavy augmentation.
", "votes": 0, "replies": [ { "id": 596775, "author_name": "abnerzhang", "author_url": "", "post_date": "2019-08-11T09:21:25.797000", "content": "Cool, let's do it
", "votes": 0, "replies": [] }, { "id": 597013, "author_name": "", "author_url": "", "post_date": "2019-08-11T16:50:30.577000", "content": "", "votes": 0, "replies": [] }, { "id": 597027, "author_name": "DeepInvolution", "author_url": "", "post_date": "2019-08-11T17:08:51.853000", "content": "I have requested to merge. Fighting!
", "votes": 0, "replies": [] }, { "id": 597221, "author_name": "abnerzhang", "author_url": "", "post_date": "2019-08-12T02:56:21.510000", "content": "what's your linkedin or QQ?
", "votes": 0, "replies": [] }, { "id": 597225, "author_name": "", "author_url": "", "post_date": "2019-08-12T03:09:18.437000", "content": "", "votes": 0, "replies": [] }, { "id": 597504, "author_name": "samshipengs", "author_url": "", "post_date": "2019-08-12T12:32:23.557000", "content": "@abnerzhang @xiaojiu1414 I'm also interested joining the team, please let me know what you think?
", "votes": 0, "replies": [] } ] }, { "id": 606443, "author_name": "", "author_url": "", "post_date": "2019-08-23T15:27:56.160000", "content": "", "votes": 0, "replies": [] }, { "id": 605910, "author_name": "", "author_url": "", "post_date": "2019-08-23T00:22:29.883000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "596063": "Follow DrHB's suggestion, use B3 net work, more on clean the 2019 data and select on 2015 data, and preprocess. one model get 0.807 with cv 0.92. If your approach is different and have similar score, maybe we could ensemble our models to get high score and generalize our approaches to avoid shakeup in private data set.", "607154": "I have one simple model based on 2019 data (LB0.818) with the new GPU restriction it will be too hard so if someone wants to team let me know", "606403": "Why there is this restriction?\n\nThe combined submission total (331) exceeds the maximum allowed (285). Unfortunately this restriction will likely apply up to the competition's team merge deadline.", "605950": "I'd like to team up. Best single model is 0.807 DenseNet with ordinal regression. ", "597063": "I'd also like to team up, single (efficientnet) model 0.79LB, 0.9CV on 2019 data only, no cleaning", "596673": "What do you mean by clean the 2019 data? ", "596650": "Hi I want to team up. I got higher CV score, but lower LB score. My highest single model is 0.799 with CV 0.90. Earger to exchange with ideas to decrease the CV gap and ensemble methods.\n\nThe best CV score is based on resnext with heavy augmentation.", "606443": "", "605910": "" } }