![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F10155416%2F608a7908f1445ebb98b474dd963245f5%2Fnewplot%20(13).png?generation=1714751598188608&alt=media\")
Since in this competition the metric is macro-R2 score, the more targets we can predict well, the better score we get (obviously :D)
\nPublic work (neural network baselines) set a very harsh filter on targets of df[target_col].std() > 1e-8, which doesn't cover all predictable columns.
I made a notebook showing a baseline R2 scores on (in my opinion) predictable columns and outlined some unpredictable targets.
\nShare your findings!
", "messages": [ { "id": 2791045, "postDate": "2024-05-03T12:55:23.907Z", "content": "Since in this competition the metric is macro-R2 score, the more targets we can predict well, the better score we get (obviously :D)
\nPublic work (neural network baselines) set a very harsh filter on targets of df[target_col].std() > 1e-8, which doesn't cover all predictable columns.
I made a notebook showing a baseline R2 scores on (in my opinion) predictable columns and outlined some unpredictable targets.
\nShare your findings!
", "rawMarkdown": "Since in this competition the metric is macro-R2 score, the more targets we can predict well, the better score we get (obviously :D) \n\nPublic work (neural network baselines) set a very harsh filter on targets of `df[target_col].std() > 1e-8`, which doesn't cover all predictable columns.\n\nI made a [notebook](https://www.kaggle.com/code/martynoveduard/how-many-targets-can-we-predict) showing a baseline R2 scores on (in my opinion) predictable columns and outlined some unpredictable targets.\n\nShare your findings! ", "votes": 15 }, { "id": 2791290, "postDate": "2024-05-03T15:21:52.127Z", "content": "Have you tried for unpredictable targets ptend_q0002_{idx} = -state_q0002_{idx}/1200?
", "rawMarkdown": "Have you tried for unpredictable targets ptend_q0002_{idx} = -state_q0002_{idx}/1200?", "votes": 16, "replies": [ { "id": 2791503, "postDate": "2024-05-03T17:30:18.487Z", "content": "Thank you for the tip, it worked (as a post-processing) :]
\nI'll try to incorporate this knowledge into the model.
", "rawMarkdown": "Thank you for the tip, it worked (as a post-processing) :]\n\nI'll try to incorporate this knowledge into the model.", "votes": 1 }, { "id": 2844973, "postDate": "2024-05-30T10:15:08.387Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 2791333, "postDate": "2024-05-03T15:55:31.303Z", "content": "
For me the \"impossible\" columns are those gaps including 72-79, 132, 133, 139-148, 192-196. 25 in total :)
\nIt looks like I nudged it down to 1e-10 for this result
Have you solved this problem?
", "rawMarkdown": "Have you solved this problem?" } ] }, { "id": 2792614, "postDate": "2024-05-04T09:44:33.590Z", "content": "It's off topic. I'm wondering how many GPUs are required for this competition.
", "rawMarkdown": "It's off topic. I'm wondering how many GPUs are required for this competition.", "votes": 2, "replies": [ { "id": 2792691, "postDate": "2024-05-04T10:37:05.057Z", "content": "Hi! I'm using single 4090
", "rawMarkdown": "Hi! I'm using single 4090", "votes": 5 }, { "id": 2793116, "postDate": "2024-05-04T14:51:24.613Z", "content": "I have single RTX 4090.
", "rawMarkdown": "I have single RTX 4090.", "votes": 1 } ] } ], "comments": [ { "id": 2791290, "author_name": "Jano123", "author_url": "", "post_date": "2024-05-03T15:21:52.127000", "content": "Have you tried for unpredictable targets ptend_q0002_{idx} = -state_q0002_{idx}/1200?
", "votes": 16, "replies": [ { "id": 2791503, "author_name": "slime", "author_url": "", "post_date": "2024-05-03T17:30:18.487000", "content": "Thank you for the tip, it worked (as a post-processing) :]
\nI'll try to incorporate this knowledge into the model.
", "votes": 1, "replies": [] }, { "id": 2844973, "author_name": "", "author_url": "", "post_date": "2024-05-30T10:15:08.387000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 2791333, "author_name": "Jekasm19", "author_url": "", "post_date": "2024-05-03T15:55:31.303000", "content": "
For me the \"impossible\" columns are those gaps including 72-79, 132, 133, 139-148, 192-196. 25 in total :)
\nIt looks like I nudged it down to 1e-10 for this result
Have you solved this problem?
", "votes": 0, "replies": [] } ] }, { "id": 2792614, "author_name": "yuanzhe zhou", "author_url": "", "post_date": "2024-05-04T09:44:33.590000", "content": "It's off topic. I'm wondering how many GPUs are required for this competition.
", "votes": 2, "replies": [ { "id": 2792691, "author_name": "slime", "author_url": "", "post_date": "2024-05-04T10:37:05.057000", "content": "Hi! I'm using single 4090
", "votes": 5, "replies": [] }, { "id": 2793116, "author_name": "milann", "author_url": "", "post_date": "2024-05-04T14:51:24.613000", "content": "I have single RTX 4090.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2791045": "Since in this competition the metric is macro-R2 score, the more targets we can predict well, the better score we get (obviously :D) \n\nPublic work (neural network baselines) set a very harsh filter on targets of `df[target_col].std() > 1e-8`, which doesn't cover all predictable columns.\n\nI made a [notebook](https://www.kaggle.com/code/martynoveduard/how-many-targets-can-we-predict) showing a baseline R2 scores on (in my opinion) predictable columns and outlined some unpredictable targets.\n\nShare your findings! ", "2791290": "Have you tried for unpredictable targets ptend_q0002_{idx} = -state_q0002_{idx}/1200?", "2791333": ".png?generation=1714751598188608&alt=media)\n\nFor me the \"impossible\" columns are those gaps including 72-79, 132, 133, 139-148, 192-196. 25 in total :) \nIt looks like I nudged it down to 1e-10 for this result", "2792614": "It's off topic. I'm wondering how many GPUs are required for this competition." } }