It seems that the best results come from modeling and not from an ML, so I wonder if this is really an ML competition.
\nThe host ( @GordonYip ) posted a very interesting starter workbook but I haven't seen a workbook based on the approach achieving a \"decent\" score or the host notebook with a submission.
Congratulations to the French team ! Space coders ( @vdebout & @skril31 ) who demonstrated their in-depth knowledge, because they are โprofessionalsโ in the space field. They achieved an incredible score and left everyone behind ๐. I'm afraid the competition is over (no ML for me) and I've published my work for those who still want to compete. Good luck and enjoy ๐
\nhttps://www.kaggle.com/code/pourchot/ariel-data-challenge-2024
It seems that the best results come from modeling and not from an ML, so I wonder if this is really an ML competition.
\nThe host ( @GordonYip ) posted a very interesting starter workbook but I haven't seen a workbook based on the approach achieving a \"decent\" score or the host notebook with a submission.
Congratulations to the French team ! Space coders ( @vdebout & @skril31 ) who demonstrated their in-depth knowledge, because they are โprofessionalsโ in the space field. They achieved an incredible score and left everyone behind ๐. I'm afraid the competition is over (no ML for me) and I've published my work for those who still want to compete. Good luck and enjoy ๐
\nhttps://www.kaggle.com/code/pourchot/ariel-data-challenge-2024
A simple 2D CNN model here with public LB 0.545 ๐
", "rawMarkdown": "A simple 2D CNN model here with public LB 0.545 ๐ ", "votes": 6, "replies": [ { "id": 2985140, "postDate": "2024-09-10T11:27:56.633Z", "content": "Hi @wuliaokaola, I'm surprised that a simple 2D CN can achieve such score, congrats !
", "rawMarkdown": "Hi @wuliaokaola, I'm surprised that a simple 2D CN can achieve such score, congrats !" }, { "id": 2985149, "postDate": "2024-09-10T11:34:29.497Z", "content": "This is impressive. I trained four models only to get a score of 0.556.
", "rawMarkdown": "This is impressive. I trained four models only to get a score of 0.556.", "votes": 3 }, { "id": 2985189, "postDate": "2024-09-10T12:14:54.800Z", "content": "Up to 0.551 by changing sigma.
", "rawMarkdown": "Up to 0.551 by changing sigma.", "votes": 2, "replies": [ { "id": 2985192, "postDate": "2024-09-10T12:19:07.023Z", "content": "Just curios: is the approach very different from my 2d cnn https://www.kaggle.com/competitions/ariel-data-challenge-2024/discussion/532740?
", "rawMarkdown": "Just curios: is the approach very different from my 2d cnn https://www.kaggle.com/competitions/ariel-data-challenge-2024/discussion/532740?", "votes": 2 }, { "id": 2985202, "postDate": "2024-09-10T12:32:23.920Z", "content": "Actually I don't know. Because I haven't yet read your notebook.
\nI use 128x288x1 as input and 283 as output. The backbone is EfficientNetV2B3.
May I recommend using MobileNet_v2? In my experiments, it was superior to EfficientNet's.
", "rawMarkdown": "May I recommend using MobileNet_v2? In my experiments, it was superior to EfficientNet's.", "votes": 3 }, { "id": 2985211, "postDate": "2024-09-10T12:50:06.237Z", "content": "Thank you. I'll try it.
", "rawMarkdown": "Thank you. I'll try it.", "votes": 1 } ] } ] }, { "id": 2985089, "postDate": "2024-09-10T10:02:32.390Z", "content": "rather signal processing methods and optimization. Models can't handle distribution shift well enough.
", "rawMarkdown": "rather signal processing methods and optimization. Models can't handle distribution shift well enough. ", "votes": 6 }, { "id": 2985106, "postDate": "2024-09-10T10:31:59.460Z", "content": "I don't know yet if this competition has a place for ML models for top scores. It may have a place mainly for predicting sigma.
\nHowever, is is definitely a data science competition :)
@pourchot you have use a ridge model in your notebook that I think is included in a private data source.
\nIs not that ML?
", "rawMarkdown": "@pourchot you have use a ridge model in your notebook that I think is included in a private data source.\n\nIs not that ML?", "replies": [ { "id": 2986145, "postDate": "2024-09-11T12:08:48.997Z", "content": "Hi @agenlu, indeed but a very simple one ๐, but my point is to achieve a score around or above 0.6
", "rawMarkdown": "Hi @agenlu, indeed but a very simple one ๐, but my point is to achieve a score around or above 0.6", "votes": 1 } ] } ], "comments": [ { "id": 2985097, "author_name": "Johnny Lee", "author_url": "", "post_date": "2024-09-10T10:14:08.440000", "content": "A simple 2D CNN model here with public LB 0.545 ๐
", "votes": 6, "replies": [ { "id": 2985140, "author_name": "Laurent Pourchot", "author_url": "", "post_date": "2024-09-10T11:27:56.633000", "content": "Hi @wuliaokaola, I'm surprised that a simple 2D CN can achieve such score, congrats !
", "votes": 0, "replies": [] }, { "id": 2985149, "author_name": "ChingYinNg", "author_url": "", "post_date": "2024-09-10T11:34:29.497000", "content": "This is impressive. I trained four models only to get a score of 0.556.
", "votes": 3, "replies": [] }, { "id": 2985189, "author_name": "Johnny Lee", "author_url": "", "post_date": "2024-09-10T12:14:54.800000", "content": "Up to 0.551 by changing sigma.
", "votes": 2, "replies": [ { "id": 2985192, "author_name": "Sergei Fironov", "author_url": "", "post_date": "2024-09-10T12:19:07.023000", "content": "Just curios: is the approach very different from my 2d cnn https://www.kaggle.com/competitions/ariel-data-challenge-2024/discussion/532740?
", "votes": 2, "replies": [] }, { "id": 2985202, "author_name": "Johnny Lee", "author_url": "", "post_date": "2024-09-10T12:32:23.920000", "content": "Actually I don't know. Because I haven't yet read your notebook.
\nI use 128x288x1 as input and 283 as output. The backbone is EfficientNetV2B3.
May I recommend using MobileNet_v2? In my experiments, it was superior to EfficientNet's.
", "votes": 3, "replies": [] }, { "id": 2985211, "author_name": "Johnny Lee", "author_url": "", "post_date": "2024-09-10T12:50:06.237000", "content": "Thank you. I'll try it.
", "votes": 1, "replies": [] } ] } ] }, { "id": 2985089, "author_name": "Sergei Fironov", "author_url": "", "post_date": "2024-09-10T10:02:32.390000", "content": "rather signal processing methods and optimization. Models can't handle distribution shift well enough.
", "votes": 6, "replies": [] }, { "id": 2985106, "author_name": "greySnow", "author_url": "", "post_date": "2024-09-10T10:31:59.460000", "content": "I don't know yet if this competition has a place for ML models for top scores. It may have a place mainly for predicting sigma.
\nHowever, is is definitely a data science competition :)
@pourchot you have use a ridge model in your notebook that I think is included in a private data source.
\nIs not that ML?
", "votes": 0, "replies": [ { "id": 2986145, "author_name": "Laurent Pourchot", "author_url": "", "post_date": "2024-09-11T12:08:48.997000", "content": "Hi @agenlu, indeed but a very simple one ๐, but my point is to achieve a score around or above 0.6
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2984894": "It seems that the best results come from modeling and not from an ML, so I wonder if this is really an ML competition.\nThe host ( @GordonYip ) posted a very interesting starter workbook but I haven't seen a workbook based on the approach achieving a \"decent\" score or the host notebook with a submission.\n\nCongratulations to the French team ! Space coders ( @vdebout & @skril31 ) who demonstrated their in-depth knowledge, because they are โprofessionalsโ in the space field. They achieved an incredible score and left everyone behind ๐. I'm afraid the competition is over (no ML for me) and I've published my work for those who still want to compete. Good luck and enjoy ๐\nhttps://www.kaggle.com/code/pourchot/ariel-data-challenge-2024", "2985097": "A simple 2D CNN model here with public LB 0.545 ๐ ", "2985089": "rather signal processing methods and optimization. Models can't handle distribution shift well enough. ", "2985106": "I don't know yet if this competition has a place for ML models for top scores. It may have a place mainly for predicting sigma.\nHowever, is *is* definitely a data science competition :)", "2986080": "@pourchot you have use a ridge model in your notebook that I think is included in a private data source.\n\nIs not that ML?" } }