{ "id": 527245, "title": "ResNet 1D results", "url": "/competitions/leash-BELKA/discussion/527245", "author_name": "", "post_date": "2024-08-11T08:02:07.713007100Z", "votes": 2, "comment_count": 2, "views": 0, "content": "

After seeing CNN 1D outperform, I tried on training ResNet 1D model (incorporating Residual blocks, 20 epochs), to my surprise, it performed well on public LB giving 0.402 (0.395 CNN 1D for 20 epochs) but could score only 0.240 on Private LB.
\nDid somebody else also try ResNet 1D?

\n

\"\"

", "messages": [ { "id": "2955501", "postDate": "08/11/2024 08:02:07", "content": "

After seeing CNN 1D outperform, I tried on training ResNet 1D model (incorporating Residual blocks, 20 epochs), to my surprise, it performed well on public LB giving 0.402 (0.395 CNN 1D for 20 epochs) but could score only 0.240 on Private LB.
\nDid somebody else also try ResNet 1D?

\n

\"\"

", "rawMarkdown": "After seeing CNN 1D outperform, I tried on training ResNet 1D model (incorporating Residual blocks, 20 epochs), to my surprise, it performed well on public LB giving 0.402 (0.395 CNN 1D for 20 epochs) but could score only 0.240 on Private LB. \nDid somebody else also try ResNet 1D? \n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2F7629adfe7d2faba3d563cf89f7406cff%2FScreen%20Shot%202024-08-11%20at%201.27.30%20PM.png?generation=1723363272258378&alt=media)", "votes": null }, { "id": "2957306", "postDate": "08/13/2024 02:45:03", "content": "

How well does it perform on the different data subsets - shared, non-shared, and non-triazine?

", "rawMarkdown": "How well does it perform on the different data subsets - shared, non-shared, and non-triazine?", "votes": null }, { "id": "2957360", "postDate": "08/13/2024 03:47:48", "content": "

Hey @kirkdco
\nThese are my results. I'd say it's performance on the non-share part for BRD4 and sEH is relatively low, lowering down the scores, for kin0 it's better than the 0.310 notebook and for share part it's kind of comparable. The reasoning needs to be understood for it's relatively concerning low performance with the two proteins and come up with a way to tackle it.
\n\"\"
\n\"\"
\n\"\"

", "rawMarkdown": "Hey @kirkdco \nThese are my results. I'd say it's performance on the non-share part for BRD4 and sEH is relatively low, lowering down the scores, for kin0 it's better than the 0.310 notebook and for share part it's kind of comparable. The reasoning needs to be understood for it's relatively concerning low performance with the two proteins and come up with a way to tackle it. \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2Fc79a5ea2eb2556a152875ec72ddc1f17%2FScreen%20Shot%202024-08-13%20at%208.56.08%20AM.png?generation=1723519770345548&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2Fdd086f41e31f701899f059070be43ce8%2FScreen%20Shot%202024-08-13%20at%208.56.55%20AM.png?generation=1723519790649185&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2F2a7bb52e64ffa208f094179f27d96353%2FScreen%20Shot%202024-08-13%20at%208.57.54%20AM.png?generation=1723519806131386&alt=media)", "votes": null } ], "comments": [ { "id": 2957306, "author_name": "kirkdco", "author_url": "", "post_date": "08/13/2024 02:45:03", "content": "

How well does it perform on the different data subsets - shared, non-shared, and non-triazine?

", "votes": null, "replies": [ { "id": 2957360, "author_name": "ahsuna123", "author_url": "", "post_date": "08/13/2024 03:47:48", "content": "

Hey @kirkdco
\nThese are my results. I'd say it's performance on the non-share part for BRD4 and sEH is relatively low, lowering down the scores, for kin0 it's better than the 0.310 notebook and for share part it's kind of comparable. The reasoning needs to be understood for it's relatively concerning low performance with the two proteins and come up with a way to tackle it.
\n\"\"
\n\"\"
\n\"\"

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2955501": "After seeing CNN 1D outperform, I tried on training ResNet 1D model (incorporating Residual blocks, 20 epochs), to my surprise, it performed well on public LB giving 0.402 (0.395 CNN 1D for 20 epochs) but could score only 0.240 on Private LB. \nDid somebody else also try ResNet 1D? \n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2F7629adfe7d2faba3d563cf89f7406cff%2FScreen%20Shot%202024-08-11%20at%201.27.30%20PM.png?generation=1723363272258378&alt=media)", "2957306": "How well does it perform on the different data subsets - shared, non-shared, and non-triazine?", "2957360": "Hey @kirkdco \nThese are my results. I'd say it's performance on the non-share part for BRD4 and sEH is relatively low, lowering down the scores, for kin0 it's better than the 0.310 notebook and for share part it's kind of comparable. The reasoning needs to be understood for it's relatively concerning low performance with the two proteins and come up with a way to tackle it. \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2Fc79a5ea2eb2556a152875ec72ddc1f17%2FScreen%20Shot%202024-08-13%20at%208.56.08%20AM.png?generation=1723519770345548&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2Fdd086f41e31f701899f059070be43ce8%2FScreen%20Shot%202024-08-13%20at%208.56.55%20AM.png?generation=1723519790649185&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2F2a7bb52e64ffa208f094179f27d96353%2FScreen%20Shot%202024-08-13%20at%208.57.54%20AM.png?generation=1723519806131386&alt=media)" }, "source": "meta" }