The competition is coming to an end pretty soon so everybody is pushing hard to improve the public leaderboard score and also the robustness of the models.\nWhat are you focusing on in this final days ? Trying new preprocessing methods, experimenting new optimizers, trying new model arhitectures, changing the entire pipeline, identifying and removing mislabeled data from the set ?\nI personally am working on tuning the best models that I got so far and trying to determine mislabeled data from the dataset.
", "messages": [ { "id": 929303, "postDate": "2020-07-14T15:41:14.963Z", "content": "The competition is coming to an end pretty soon so everybody is pushing hard to improve the public leaderboard score and also the robustness of the models.\nWhat are you focusing on in this final days ? Trying new preprocessing methods, experimenting new optimizers, trying new model arhitectures, changing the entire pipeline, identifying and removing mislabeled data from the set ?\nI personally am working on tuning the best models that I got so far and trying to determine mislabeled data from the dataset.
", "rawMarkdown": "The competition is coming to an end pretty soon so everybody is pushing hard to improve the public leaderboard score and also the robustness of the models.\nWhat are you focusing on in this final days ? Trying new preprocessing methods, experimenting new optimizers, trying new model arhitectures, changing the entire pipeline, identifying and removing mislabeled data from the set ?\nI personally am working on tuning the best models that I got so far and trying to determine mislabeled data from the dataset. ", "votes": 13 }, { "id": 929503, "postDate": "2020-07-14T17:56:06.060Z", "content": "Our team is going to try optimize seed while making cute PANDA gifs =) Good Luck! =)
", "rawMarkdown": "Our team is going to try optimize seed while making cute PANDA gifs =) Good Luck! =) ", "votes": 3, "replies": [ { "id": 929578, "postDate": "2020-07-14T18:58:23.623Z", "content": "haha, Good Luck guys !
", "rawMarkdown": "haha, Good Luck guys !", "votes": 1 } ] }, { "id": 929717, "postDate": "2020-07-14T22:21:04.957Z", "content": "Speeding up the inference.\nAs I experimented out of Kaggle (offline), now I need to make the inference work fast inside Kaggle submission docker container.
", "rawMarkdown": "Speeding up the inference.\nAs I experimented out of Kaggle (offline), now I need to make the inference work fast inside Kaggle submission docker container.", "votes": 1, "replies": [ { "id": 930137, "postDate": "2020-07-15T08:15:53.827Z", "content": "Good Luck @dgrechka
", "rawMarkdown": "Good Luck @dgrechka " } ] }, { "id": 931722, "postDate": "2020-07-16T12:06:22.670Z", "content": "thinking about tiling strategy. I've searched papers, tried mentioned preprocessing methods but none working.
", "rawMarkdown": "thinking about tiling strategy. I've searched papers, tried mentioned preprocessing methods but none working. ", "isDeleted": true, "replies": [ { "id": 936464, "postDate": "2020-07-20T08:34:34.837Z", "content": "@muerbingsha What tile method did you use for your best score ?
", "rawMarkdown": "@muerbingsha What tile method did you use for your best score ?" }, { "id": 936540, "postDate": "2020-07-20T09:57:31.023Z", "content": "still 256X256X36 works better. how's yours. Your rank is high.
", "rawMarkdown": "still 256X256X36 works better. how's yours. Your rank is high.", "isDeleted": true }, { "id": 936683, "postDate": "2020-07-20T12:48:07.437Z", "content": "256X256X36 also
", "rawMarkdown": "256X256X36 also" }, { "id": 936962, "postDate": "2020-07-20T16:17:47.637Z", "content": "Hi @vladvdv and @muerbingsha for me the following combination works best. \n256x256x48 for efficient-net-b0\n256x256x34 for efficient-net-b1/b2 \nfor inference i am taking an average of the three models. still looking for better ways to ensemble the models.
", "rawMarkdown": "Hi @vladvdv and @muerbingsha for me the following combination works best. \n256x256x48 for efficient-net-b0\n256x256x34 for efficient-net-b1/b2 \nfor inference i am taking an average of the three models. still looking for better ways to ensemble the models. \n \n" } ] }, { "id": 929607, "postDate": "2020-07-14T19:28:06.143Z", "rawMarkdown": "", "votes": 1, "isDeleted": true, "replies": [ { "id": 931443, "postDate": "2020-07-16T07:44:31.403Z", "content": "@anthonymcqueen Did you finally succeed ?
", "rawMarkdown": "@anthonymcqueen Did you finally succeed ?" } ] } ], "comments": [ { "id": 929503, "author_name": "DrHB", "author_url": "", "post_date": "2020-07-14T17:56:06.060000", "content": "Our team is going to try optimize seed while making cute PANDA gifs =) Good Luck! =)
", "votes": 3, "replies": [ { "id": 929578, "author_name": "Vlad Vaduva", "author_url": "", "post_date": "2020-07-14T18:58:23.623000", "content": "haha, Good Luck guys !
", "votes": 1, "replies": [] } ] }, { "id": 929717, "author_name": "Dmitry A. Grechka", "author_url": "", "post_date": "2020-07-14T22:21:04.957000", "content": "Speeding up the inference.\nAs I experimented out of Kaggle (offline), now I need to make the inference work fast inside Kaggle submission docker container.
", "votes": 1, "replies": [ { "id": 930137, "author_name": "Vlad Vaduva", "author_url": "", "post_date": "2020-07-15T08:15:53.827000", "content": "Good Luck @dgrechka
", "votes": 0, "replies": [] } ] }, { "id": 931722, "author_name": "", "author_url": "", "post_date": "2020-07-16T12:06:22.670000", "content": "thinking about tiling strategy. I've searched papers, tried mentioned preprocessing methods but none working.
", "votes": 0, "replies": [ { "id": 936464, "author_name": "Vlad Vaduva", "author_url": "", "post_date": "2020-07-20T08:34:34.837000", "content": "@muerbingsha What tile method did you use for your best score ?
", "votes": 0, "replies": [] }, { "id": 936540, "author_name": "", "author_url": "", "post_date": "2020-07-20T09:57:31.023000", "content": "still 256X256X36 works better. how's yours. Your rank is high.
", "votes": 0, "replies": [] }, { "id": 936683, "author_name": "Vlad Vaduva", "author_url": "", "post_date": "2020-07-20T12:48:07.437000", "content": "256X256X36 also
", "votes": 0, "replies": [] }, { "id": 936962, "author_name": "Mehul Sampat", "author_url": "", "post_date": "2020-07-20T16:17:47.637000", "content": "Hi @vladvdv and @muerbingsha for me the following combination works best. \n256x256x48 for efficient-net-b0\n256x256x34 for efficient-net-b1/b2 \nfor inference i am taking an average of the three models. still looking for better ways to ensemble the models.
", "votes": 0, "replies": [] } ] }, { "id": 929607, "author_name": "", "author_url": "", "post_date": "2020-07-14T19:28:06.143000", "content": "", "votes": 1, "replies": [ { "id": 931443, "author_name": "Vlad Vaduva", "author_url": "", "post_date": "2020-07-16T07:44:31.403000", "content": "@anthonymcqueen Did you finally succeed ?
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "929303": "The competition is coming to an end pretty soon so everybody is pushing hard to improve the public leaderboard score and also the robustness of the models.\nWhat are you focusing on in this final days ? Trying new preprocessing methods, experimenting new optimizers, trying new model arhitectures, changing the entire pipeline, identifying and removing mislabeled data from the set ?\nI personally am working on tuning the best models that I got so far and trying to determine mislabeled data from the dataset. ", "929503": "Our team is going to try optimize seed while making cute PANDA gifs =) Good Luck! =) ", "929717": "Speeding up the inference.\nAs I experimented out of Kaggle (offline), now I need to make the inference work fast inside Kaggle submission docker container.", "931722": "thinking about tiling strategy. I've searched papers, tried mentioned preprocessing methods but none working. ", "929607": "" } }