Could someone please explain why a new sample_submission was updated, and it appears that the weights inside are only 0 and 1? How does this affect our competition?
", "messages": [ { "id": 2879505, "postDate": "2024-06-19T16:08:35.990Z", "content": "Could someone please explain why a new sample_submission was updated, and it appears that the weights inside are only 0 and 1? How does this affect our competition?
", "rawMarkdown": "Could someone please explain why a new sample_submission was updated, and it appears that the weights inside are only 0 and 1? How does this affect our competition?", "votes": 1 }, { "id": 2888890, "postDate": "2024-06-25T05:52:39.477Z", "content": "Let me take a look at the answer content
", "rawMarkdown": "Let me take a look at the answer content\n\n" }, { "id": 2879985, "postDate": "2024-06-20T00:56:50.980Z", "content": "\nIf you multiplied by the weights before model fit you may have some issues with float32 vs float64. The old weights minimized a few of the issues using float32.
", "rawMarkdown": "[Read](https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/513193)\n\nIf you multiplied by the weights before model fit you may have some issues with float32 vs float64. The old weights minimized a few of the issues using float32.", "replies": [ { "id": 2881087, "postDate": "2024-06-20T14:26:45.640Z", "content": "Thank you very much
", "rawMarkdown": "Thank you very much" } ] }, { "id": 2879543, "postDate": "2024-06-19T16:47:01.867Z", "content": "Previously, I was using a PyTorch baseline that predicted data multiplied by a weight. Some targets originally had a weight of zero, thus the predictions were zero. However, now these targets, which initially had a weight of zero, have non-zero weights. How should this be resolved? Re-training seems to be more labor-intensive.
", "rawMarkdown": "Previously, I was using a PyTorch baseline that predicted data multiplied by a weight. Some targets originally had a weight of zero, thus the predictions were zero. However, now these targets, which initially had a weight of zero, have non-zero weights. How should this be resolved? Re-training seems to be more labor-intensive.", "replies": [ { "id": 2881370, "postDate": "2024-06-20T16:59:19.887Z", "content": "this solved it for me: https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/513220#2880523
", "rawMarkdown": "this solved it for me: https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/513220#2880523", "votes": 1 }, { "id": 2881371, "postDate": "2024-06-20T16:59:58.587Z", "content": "More info: https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/499896#2791290
", "rawMarkdown": "More info: https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/499896#2791290" } ] } ], "comments": [ { "id": 2888890, "author_name": "iKunKunKunKun", "author_url": "", "post_date": "2024-06-25T05:52:39.477000", "content": "Let me take a look at the answer content
", "votes": 0, "replies": [] }, { "id": 2879985, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2024-06-20T00:56:50.980000", "content": "\nIf you multiplied by the weights before model fit you may have some issues with float32 vs float64. The old weights minimized a few of the issues using float32.
", "votes": 0, "replies": [ { "id": 2881087, "author_name": "Chengwei Yan", "author_url": "", "post_date": "2024-06-20T14:26:45.640000", "content": "Thank you very much
", "votes": 0, "replies": [] } ] }, { "id": 2879543, "author_name": "Chengwei Yan", "author_url": "", "post_date": "2024-06-19T16:47:01.867000", "content": "Previously, I was using a PyTorch baseline that predicted data multiplied by a weight. Some targets originally had a weight of zero, thus the predictions were zero. However, now these targets, which initially had a weight of zero, have non-zero weights. How should this be resolved? Re-training seems to be more labor-intensive.
", "votes": 0, "replies": [ { "id": 2881370, "author_name": "Juan D C F", "author_url": "", "post_date": "2024-06-20T16:59:19.887000", "content": "this solved it for me: https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/513220#2880523
", "votes": 1, "replies": [] }, { "id": 2881371, "author_name": "Juan D C F", "author_url": "", "post_date": "2024-06-20T16:59:58.587000", "content": "More info: https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/499896#2791290
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2879505": "Could someone please explain why a new sample_submission was updated, and it appears that the weights inside are only 0 and 1? How does this affect our competition?", "2888890": "Let me take a look at the answer content\n\n", "2879985": "[Read](https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/513193)\n\nIf you multiplied by the weights before model fit you may have some issues with float32 vs float64. The old weights minimized a few of the issues using float32.", "2879543": "Previously, I was using a PyTorch baseline that predicted data multiplied by a weight. Some targets originally had a weight of zero, thus the predictions were zero. However, now these targets, which initially had a weight of zero, have non-zero weights. How should this be resolved? Re-training seems to be more labor-intensive." } }