I've seen from @coolz a really cool 1d implementation from a previous competition https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/overview . I was wondering if anybody had success with it this comp?
", "messages": [ { "id": "2825799", "postDate": "05/20/2024 15:14:16", "content": "I've seen from @coolz a really cool 1d implementation from a previous competition https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/overview . I was wondering if anybody had success with it this comp?
", "rawMarkdown": "I've seen from @coolz a really cool 1d implementation from a previous competition https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/overview . I was wondering if anybody had success with it this comp?", "votes": null }, { "id": "2826343", "postDate": "05/20/2024 21:03:35", "content": "Well, maybe he is using it here.
", "rawMarkdown": "Well, maybe he is using it here.", "votes": null }, { "id": "2826402", "postDate": "05/20/2024 22:28:24", "content": "I have
\n1D CNN with 0.66 LB
\n1D CNN followed by RNN 0.66 LB
\nTransformer Architecture with 0.66 LB
\n2D CNN with 0.67 LB
Didn't find time to research more, but a lot of possibilities.
", "rawMarkdown": "I have \n1D CNN with 0.66 LB\n1D CNN followed by RNN 0.66 LB\nTransformer Architecture with 0.66 LB\n2D CNN with 0.67 LB\n\nDidn't find time to research more, but a lot of possibilities.", "votes": null }, { "id": "2826488", "postDate": "05/21/2024 01:17:54", "content": "the only way to know is to try haha! I havent seen any discussion with anybody using 1d waves, and thought I would shoot their experiences. Ya got any tips for us who started a lil late :)
", "rawMarkdown": "the only way to know is to try haha! I havent seen any discussion with anybody using 1d waves, and thought I would shoot their experiences. Ya got any tips for us who started a lil late :)", "votes": null }, { "id": "2826489", "postDate": "05/21/2024 01:18:06", "content": "Thank you for sharing :)
", "rawMarkdown": "Thank you for sharing :)", "votes": null }, { "id": "2826632", "postDate": "05/21/2024 04:00:17", "content": "so many 0.66
", "rawMarkdown": "so many 0.66", "votes": null }, { "id": "2827769", "postDate": "05/21/2024 16:24:01", "content": "I'm so glad you brought this up. I'm hoping to see someone using 1-D in the top of the final leaderboard. My very very initial look at this, I was trying to use wav2vec2 on the unlabeled soundscapes. Intuition being if you build a solid base from those (I think) more test file like audio samples it could help during final inference. I got as far as using the Hugging Face tutorial and saw the masked loss consistently reduce, but I didn't do nearly enough to verify that this works for classification.
\nHopefully you can find some cool ideas to make 1D work.
", "rawMarkdown": "I'm so glad you brought this up. I'm hoping to see someone using 1-D in the top of the final leaderboard. My very very initial look at this, I was trying to use wav2vec2 on the unlabeled soundscapes. Intuition being if you build a solid base from those (I think) more test file like audio samples it could help during final inference. I got as far as using the Hugging Face tutorial and saw the masked loss consistently reduce, but I didn't do nearly enough to verify that this works for classification.\n\nHopefully you can find some cool ideas to make 1D work.", "votes": null }, { "id": "2829655", "postDate": "05/22/2024 18:02:37", "content": "Same I started a month late haha!
", "rawMarkdown": "Same I started a month late haha!", "votes": null } ], "comments": [ { "id": 2826343, "author_name": "cpmpml", "author_url": "", "post_date": "05/20/2024 21:03:35", "content": "Well, maybe he is using it here.
", "votes": null, "replies": [ { "id": 2826488, "author_name": "tashin47", "author_url": "", "post_date": "05/21/2024 01:17:54", "content": "the only way to know is to try haha! I havent seen any discussion with anybody using 1d waves, and thought I would shoot their experiences. Ya got any tips for us who started a lil late :)
", "votes": null, "replies": [ { "id": 2829655, "author_name": "max1mum", "author_url": "", "post_date": "05/22/2024 18:02:37", "content": "Same I started a month late haha!
", "votes": null, "replies": [] } ] } ] }, { "id": 2826402, "author_name": "salmanahmedtamu", "author_url": "", "post_date": "05/20/2024 22:28:24", "content": "I have
\n1D CNN with 0.66 LB
\n1D CNN followed by RNN 0.66 LB
\nTransformer Architecture with 0.66 LB
\n2D CNN with 0.67 LB
Didn't find time to research more, but a lot of possibilities.
", "votes": null, "replies": [ { "id": 2826489, "author_name": "tashin47", "author_url": "", "post_date": "05/21/2024 01:18:06", "content": "Thank you for sharing :)
", "votes": null, "replies": [] }, { "id": 2826632, "author_name": "yuanzhezhou", "author_url": "", "post_date": "05/21/2024 04:00:17", "content": "so many 0.66
", "votes": null, "replies": [] } ] }, { "id": 2827769, "author_name": "msthil", "author_url": "", "post_date": "05/21/2024 16:24:01", "content": "I'm so glad you brought this up. I'm hoping to see someone using 1-D in the top of the final leaderboard. My very very initial look at this, I was trying to use wav2vec2 on the unlabeled soundscapes. Intuition being if you build a solid base from those (I think) more test file like audio samples it could help during final inference. I got as far as using the Hugging Face tutorial and saw the masked loss consistently reduce, but I didn't do nearly enough to verify that this works for classification.
\nHopefully you can find some cool ideas to make 1D work.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2825799": "I've seen from @coolz a really cool 1d implementation from a previous competition https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/overview . I was wondering if anybody had success with it this comp?", "2826343": "Well, maybe he is using it here.", "2826402": "I have \n1D CNN with 0.66 LB\n1D CNN followed by RNN 0.66 LB\nTransformer Architecture with 0.66 LB\n2D CNN with 0.67 LB\n\nDidn't find time to research more, but a lot of possibilities.", "2826488": "the only way to know is to try haha! I havent seen any discussion with anybody using 1d waves, and thought I would shoot their experiences. Ya got any tips for us who started a lil late :)", "2826489": "Thank you for sharing :)", "2826632": "so many 0.66", "2827769": "I'm so glad you brought this up. I'm hoping to see someone using 1-D in the top of the final leaderboard. My very very initial look at this, I was trying to use wav2vec2 on the unlabeled soundscapes. Intuition being if you build a solid base from those (I think) more test file like audio samples it could help during final inference. I got as far as using the Hugging Face tutorial and saw the masked loss consistently reduce, but I didn't do nearly enough to verify that this works for classification.\n\nHopefully you can find some cool ideas to make 1D work.", "2829655": "Same I started a month late haha!" }, "source": "meta" }