After some weeks, the LB not much change. Can good team share there solution for all to build best model. It's for learn and knowledge
", "messages": [ { "id": 962792, "postDate": "2020-08-08T12:51:57.063Z", "content": "I shared Introduction to Sound Event Detection kernel just now.
\n\nI just want to show I'm working on this line and it may work (if things are organized well and trained properly), so I also put inference code on this.
\n\nHowever, I observed several people have concern on oversharing of the top solutions during the competition, and I do respect those people who have worked hard to make good model, therefore I didn't share the trained weight nor the method I trained that weight.
\n\nHope this stimulates the community and lead us to new progresses while not discouraging top competitors who have already worked hard in this competition.
", "rawMarkdown": "I shared [Introduction to Sound Event Detection](https://www.kaggle.com/hidehisaarai1213/introduction-to-sound-event-detection?scriptVersionId=40372870) kernel just now.\n\nI just want to show I'm working on this line and it may work (if things are organized well and trained properly), so I also put inference code on this.\n\nHowever, I observed several people have concern on oversharing of the top solutions during the competition, and I do respect those people who have worked hard to make good model, therefore I didn't share the trained weight nor the method I trained that weight.\n\nHope this stimulates the community and lead us to new progresses while not discouraging top competitors who have already worked hard in this competition.", "votes": 16, "replies": [ { "id": 962990, "postDate": "2020-08-08T15:30:44.567Z", "content": "thanks very much for that contribution to society.
", "rawMarkdown": "thanks very much for that contribution to society. ", "votes": -1 }, { "id": 963973, "postDate": "2020-08-09T13:34:19.703Z", "content": "Thanks for sharing your approach, hopefully people will learn a lot from it.\nMy team certainly will as we didn't use SED.
\n\nEDIT : Nevermind the weights are not public :)
", "rawMarkdown": "Thanks for sharing your approach, hopefully people will learn a lot from it.\nMy team certainly will as we didn't use SED.\n\nEDIT : Nevermind the weights are not public :) ", "votes": 2 } ] }, { "id": 961496, "postDate": "2020-08-07T08:27:28.643Z", "content": "You can find some hints in these two threads : \n- https://www.kaggle.com/c/birdsong-recognition/discussion/170821\n- https://www.kaggle.com/c/birdsong-recognition/discussion/168657
\n\nRead the comments as well, there's some interesting ideas.
\n\nI see your team is already at 0.572, you probably figured out how to improve the baseline and can already share if you feel like it. There is not a big gap between a 0.572 solution and a 0.58 solution.
\n\nOn my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that here. People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea.
\n\nLearning and knowledge are acquired when testing things, it does not necesseraly implies that these things have to work.
", "rawMarkdown": "You can find some hints in these two threads : \n- https://www.kaggle.com/c/birdsong-recognition/discussion/170821\n- https://www.kaggle.com/c/birdsong-recognition/discussion/168657\n\nRead the comments as well, there's some interesting ideas.\n\nI see your team is already at 0.572, you probably figured out how to improve the baseline and can already share if you feel like it. There is not a big gap between a 0.572 solution and a 0.58 solution.\n \nOn my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that [here](https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/151084). People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea. \n\nLearning and knowledge are acquired when testing things, it does not necesseraly implies that these things have to work.", "votes": 12 }, { "id": 958067, "postDate": "2020-08-04T18:35:13.410Z", "content": "After some weeks, the LB not much change. Can good team share there solution for all to build best model. It's for learn and knowledge
", "rawMarkdown": "After some weeks, the LB not much change. Can good team share there solution for all to build best model. It's for learn and knowledge ", "votes": 10 }, { "id": 961550, "postDate": "2020-08-07T09:24:38.130Z", "content": "I don't use anything special. If necessary, I can write my solution, but I agree with Theo Viel
\n\n\n", "rawMarkdown": "I don't use anything special. If necessary, I can write my solution, but I agree with Theo Viel\n> On my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that here. People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea.", "votes": 7 }, { "id": 961260, "postDate": "2020-08-07T03:47:50.110Z", "content": "On my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that here. People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea.
\n
If half of the team above 0.568 share (part of) successful attempt, I will share what I have done to get ~0.575.
", "rawMarkdown": "If half of the team above 0.568 share (part of) successful attempt, I will share what I have done to get ~0.575.", "votes": 4, "replies": [ { "id": 961422, "postDate": "2020-08-07T07:18:22.097Z", "content": "\n\n\nhalf of the team above 0.568 share (part of) successful attempt,
\n
Or just announce to share in discussion and share them later.
", "rawMarkdown": "> half of the team above 0.568 share (part of) successful attempt,\n\nOr just announce to share in discussion and share them later.", "votes": 1 }, { "id": 963353, "postDate": "2020-08-09T00:22:06.690Z", "content": "I think everyone is wallowing in the noise near all 'nocall's. It's a daunting task, especially with the hidden test set.
", "rawMarkdown": "I think everyone is wallowing in the noise near all 'nocall's. It's a daunting task, especially with the hidden test set." } ] }, { "id": 959454, "postDate": "2020-08-05T15:42:07.673Z", "content": "It's probably becaues model training takes quite a time. For example, training of the best public model based on ResNeSt 50 takes ~8 hours.
", "rawMarkdown": "It's probably becaues model training takes quite a time. For example, training of the best public model based on ResNeSt 50 takes ~8 hours." }, { "id": 962773, "postDate": "2020-08-08T12:36:44.370Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 962792, "author_name": "Hidehisa Arai", "author_url": "", "post_date": "2020-08-08T12:51:57.063000", "content": "I shared Introduction to Sound Event Detection kernel just now.
\n\nI just want to show I'm working on this line and it may work (if things are organized well and trained properly), so I also put inference code on this.
\n\nHowever, I observed several people have concern on oversharing of the top solutions during the competition, and I do respect those people who have worked hard to make good model, therefore I didn't share the trained weight nor the method I trained that weight.
\n\nHope this stimulates the community and lead us to new progresses while not discouraging top competitors who have already worked hard in this competition.
", "votes": 16, "replies": [ { "id": 962990, "author_name": "Manh Lab", "author_url": "", "post_date": "2020-08-08T15:30:44.567000", "content": "thanks very much for that contribution to society.
", "votes": -1, "replies": [] }, { "id": 963973, "author_name": "Theo Viel", "author_url": "", "post_date": "2020-08-09T13:34:19.703000", "content": "Thanks for sharing your approach, hopefully people will learn a lot from it.\nMy team certainly will as we didn't use SED.
\n\nEDIT : Nevermind the weights are not public :)
", "votes": 2, "replies": [] } ] }, { "id": 961496, "author_name": "Theo Viel", "author_url": "", "post_date": "2020-08-07T08:27:28.643000", "content": "You can find some hints in these two threads : \n- https://www.kaggle.com/c/birdsong-recognition/discussion/170821\n- https://www.kaggle.com/c/birdsong-recognition/discussion/168657
\n\nRead the comments as well, there's some interesting ideas.
\n\nI see your team is already at 0.572, you probably figured out how to improve the baseline and can already share if you feel like it. There is not a big gap between a 0.572 solution and a 0.58 solution.
\n\nOn my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that here. People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea.
\n\nLearning and knowledge are acquired when testing things, it does not necesseraly implies that these things have to work.
", "votes": 12, "replies": [] }, { "id": 961550, "author_name": "Kramarenko Vladislav", "author_url": "", "post_date": "2020-08-07T09:24:38.130000", "content": "I don't use anything special. If necessary, I can write my solution, but I agree with Theo Viel
\n\n\n", "votes": 7, "replies": [] }, { "id": 961260, "author_name": "Hidehisa Arai", "author_url": "", "post_date": "2020-08-07T03:47:50.110000", "content": "On my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that here. People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea.
\n
If half of the team above 0.568 share (part of) successful attempt, I will share what I have done to get ~0.575.
", "votes": 4, "replies": [ { "id": 961422, "author_name": "Hidehisa Arai", "author_url": "", "post_date": "2020-08-07T07:18:22.097000", "content": "\n\n\nhalf of the team above 0.568 share (part of) successful attempt,
\n
Or just announce to share in discussion and share them later.
", "votes": 1, "replies": [] }, { "id": 963353, "author_name": "Mark Eckdahl", "author_url": "", "post_date": "2020-08-09T00:22:06.690000", "content": "I think everyone is wallowing in the noise near all 'nocall's. It's a daunting task, especially with the hidden test set.
", "votes": 0, "replies": [] } ] }, { "id": 959454, "author_name": "Araik Tamazian", "author_url": "", "post_date": "2020-08-05T15:42:07.673000", "content": "It's probably becaues model training takes quite a time. For example, training of the best public model based on ResNeSt 50 takes ~8 hours.
", "votes": 0, "replies": [] }, { "id": 962773, "author_name": "", "author_url": "", "post_date": "2020-08-08T12:36:44.370000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "962792": "I shared [Introduction to Sound Event Detection](https://www.kaggle.com/hidehisaarai1213/introduction-to-sound-event-detection?scriptVersionId=40372870) kernel just now.\n\nI just want to show I'm working on this line and it may work (if things are organized well and trained properly), so I also put inference code on this.\n\nHowever, I observed several people have concern on oversharing of the top solutions during the competition, and I do respect those people who have worked hard to make good model, therefore I didn't share the trained weight nor the method I trained that weight.\n\nHope this stimulates the community and lead us to new progresses while not discouraging top competitors who have already worked hard in this competition.", "961496": "You can find some hints in these two threads : \n- https://www.kaggle.com/c/birdsong-recognition/discussion/170821\n- https://www.kaggle.com/c/birdsong-recognition/discussion/168657\n\nRead the comments as well, there's some interesting ideas.\n\nI see your team is already at 0.572, you probably figured out how to improve the baseline and can already share if you feel like it. There is not a big gap between a 0.572 solution and a 0.58 solution.\n \nOn my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that [here](https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/151084). People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea. \n\nLearning and knowledge are acquired when testing things, it does not necesseraly implies that these things have to work.", "958067": "After some weeks, the LB not much change. Can good team share there solution for all to build best model. It's for learn and knowledge ", "961550": "I don't use anything special. If necessary, I can write my solution, but I agree with Theo Viel\n> On my side, I don't think it's a good idea for top teams to \"overshare\" their solution. I already wrote about that here. People have worked hard to reach 0.57+ and it can be really frustrating to see everyone beating you without effort because someone shared a winning idea.", "961260": "If half of the team above 0.568 share (part of) successful attempt, I will share what I have done to get ~0.575.", "959454": "It's probably becaues model training takes quite a time. For example, training of the best public model based on ResNeSt 50 takes ~8 hours.", "962773": "" } }