![\"image\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6426215%2F5fd3fc9904fc7ccb0bb5f776241db380%2FScreenshot%202023-11-26%20at%207.57.43PM%20copy.png?generation=1701158604924450&alt=media\")
I could not do better than 0.80 mrrse in a 5-fold cross validation.
\nAm I using the wrong model architectures?
", "messages": [ { "id": "2539271", "postDate": "11/26/2023 23:27:36", "content": "I could not do better than 0.80 mrrse in a 5-fold cross validation.
\nAm I using the wrong model architectures?
", "rawMarkdown": "I could not do better than 0.80 mrrse in a 5-fold cross validation.\n\nAm I using the wrong model architectures?", "votes": null }, { "id": "2540687", "postDate": "11/27/2023 22:50:46", "content": "How do you split your data?
\nI use NK cells for validation, and the MRRMSE score is 1.1465.
NK cells are not part of the private leaderboard. Am I missing something? I split the data with sk-learn train_test_split, test size 0.20
", "rawMarkdown": "NK cells are not part of the private leaderboard. Am I missing something? I split the data with sk-learn train_test_split, test size 0.20", "votes": null }, { "id": "2540693", "postDate": "11/27/2023 23:09:36", "content": "B-cells and Myelod cells are in the private data, unless I am missing something:)
", "rawMarkdown": "B-cells and Myelod cells are in the private data, unless I am missing something:)", "votes": null }, { "id": "2540694", "postDate": "11/27/2023 23:12:08", "content": "Btw 0.80 is not observed accross all folds. Only best folds achieve this
", "rawMarkdown": "Btw 0.80 is not observed accross all folds. Only best folds achieve this", "votes": null }, { "id": "2540709", "postDate": "11/27/2023 23:35:02", "content": "It is correct that NK cells are not part of the private leaderboard. However, I am trying to get the CV score as accurate as possible by training the model without using NK cells for training, and then obtaining the CV score using NK cells.
", "rawMarkdown": "It is correct that NK cells are not part of the private leaderboard. However, I am trying to get the CV score as accurate as possible by training the model without using NK cells for training, and then obtaining the CV score using NK cells.", "votes": null }, { "id": "2541055", "postDate": "11/28/2023 07:46:50", "content": "Great point. We also see that nk cells seems to be close to LB, but not quite sure. What your ideas? how do you come up with that validation scheme?
", "rawMarkdown": "Great point. We also see that nk cells seems to be close to LB, but not quite sure. What your ideas? how do you come up with that validation scheme?", "votes": null }, { "id": "2541075", "postDate": "11/28/2023 08:04:49", "content": "I plot each cell_type genes, and I chose NK cells because the data distribution was similar to targets.
\n
Cool ! Thanks for sharing
", "rawMarkdown": "Cool ! Thanks for sharing", "votes": null } ], "comments": [ { "id": 2540687, "author_name": "masaishi", "author_url": "", "post_date": "11/27/2023 22:50:46", "content": "How do you split your data?
\nI use NK cells for validation, and the MRRMSE score is 1.1465.
NK cells are not part of the private leaderboard. Am I missing something? I split the data with sk-learn train_test_split, test size 0.20
", "votes": null, "replies": [ { "id": 2540693, "author_name": "jeannkouagou", "author_url": "", "post_date": "11/27/2023 23:09:36", "content": "B-cells and Myelod cells are in the private data, unless I am missing something:)
", "votes": null, "replies": [ { "id": 2540709, "author_name": "masaishi", "author_url": "", "post_date": "11/27/2023 23:35:02", "content": "It is correct that NK cells are not part of the private leaderboard. However, I am trying to get the CV score as accurate as possible by training the model without using NK cells for training, and then obtaining the CV score using NK cells.
", "votes": null, "replies": [] } ] } ] }, { "id": 2540694, "author_name": "jeannkouagou", "author_url": "", "post_date": "11/27/2023 23:12:08", "content": "Btw 0.80 is not observed accross all folds. Only best folds achieve this
", "votes": null, "replies": [] }, { "id": 2541055, "author_name": "alexandervc", "author_url": "", "post_date": "11/28/2023 07:46:50", "content": "Great point. We also see that nk cells seems to be close to LB, but not quite sure. What your ideas? how do you come up with that validation scheme?
", "votes": null, "replies": [ { "id": 2541075, "author_name": "masaishi", "author_url": "", "post_date": "11/28/2023 08:04:49", "content": "I plot each cell_type genes, and I chose NK cells because the data distribution was similar to targets.
\n
Cool ! Thanks for sharing
", "votes": null, "replies": [] } ] } ] } ] } ], "raw_markdown_by_id": { "2539271": "I could not do better than 0.80 mrrse in a 5-fold cross validation.\n\nAm I using the wrong model architectures?", "2540687": "How do you split your data?\nI use NK cells for validation, and the MRRMSE score is 1.1465.", "2540692": "NK cells are not part of the private leaderboard. Am I missing something? I split the data with sk-learn train_test_split, test size 0.20", "2540693": "B-cells and Myelod cells are in the private data, unless I am missing something:)", "2540694": "Btw 0.80 is not observed accross all folds. Only best folds achieve this", "2540709": "It is correct that NK cells are not part of the private leaderboard. However, I am trying to get the CV score as accurate as possible by training the model without using NK cells for training, and then obtaining the CV score using NK cells.", "2541055": "Great point. We also see that nk cells seems to be close to LB, but not quite sure. What your ideas? how do you come up with that validation scheme?", "2541075": "I plot each cell_type genes, and I chose NK cells because the data distribution was similar to targets.\n\n", "2541142": "Cool ! Thanks for sharing" }, "source": "meta" }