Could you pls share what CV and leaderboard scores you achieved using handcrafted features and GBM, both with only IMU data and with all available data?
\nI wonder how large the gap is between best gbm and neural networks on this task.
\nHere are my scores using 5 folds StratifiedGroupKFold:
\nCould you pls share what CV and leaderboard scores you achieved using handcrafted features and GBM, both with only IMU data and with all available data?
\nI wonder how large the gap is between best gbm and neural networks on this task.
\nHere are my scores using 5 folds StratifiedGroupKFold:
\nCheck this post CV-LB thread
", "rawMarkdown": "Check this post [CV-LB thread](https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/583023)", "votes": -2, "replies": [ { "id": 3241863, "postDate": "2025-07-05T09:39:06.300Z", "content": "That's a different topic.
", "rawMarkdown": "That's a different topic." } ] }, { "id": 3243074, "postDate": "2025-07-06T18:01:00.287Z", "content": "I have improved :) 0.83 cv (gbm all data) 0.75 (gbm only IMU) Public LB: 0.78
", "rawMarkdown": "I have improved :) 0.83 cv (gbm all data) 0.75 (gbm only IMU) Public LB: 0.78", "replies": [ { "id": 3243181, "postDate": "2025-07-06T20:59:16.627Z", "content": "Is that 0.78 on the public LB (and 0.75 on CV) using only IMU data, or all sensor data?
", "rawMarkdown": "Is that 0.78 on the public LB (and 0.75 on CV) using only IMU data, or all sensor data?", "votes": 1, "replies": [ { "id": 3243420, "postDate": "2025-07-07T05:59:15.770Z", "content": "It's both. You can check if nan's are present in your sequences.
", "rawMarkdown": "It's both. You can check if nan's are present in your sequences." } ] } ] }, { "id": 3241865, "postDate": "2025-07-05T09:40:55.727Z", "content": "I wonder where I'm missing out to achieve 0.77 on IMU only data.
", "rawMarkdown": "I wonder where I'm missing out to achieve 0.77 on IMU only data." }, { "id": 3241864, "postDate": "2025-07-05T09:40:18.187Z", "content": "IMU only 0.73 cv all data 0.82 cv. Public lb 0.77
", "rawMarkdown": "IMU only 0.73 cv all data 0.82 cv. Public lb 0.77" }, { "id": 3241596, "postDate": "2025-07-05T00:33:30.543Z", "content": "is there a reason to use GBM over 1d cnn?
", "rawMarkdown": "is there a reason to use GBM over 1d cnn?", "replies": [ { "id": 3242128, "postDate": "2025-07-05T15:43:46.770Z", "content": "No particular reason — mostly just scientific curiosity :)
", "rawMarkdown": "No particular reason — mostly just scientific curiosity :)", "votes": 1 } ] }, { "id": 3241207, "postDate": "2025-07-04T16:22:25.570Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 3242125, "postDate": "2025-07-05T15:42:56.110Z", "content": "Wow, great result! Have you submitted it to the leaderboard?
", "rawMarkdown": "Wow, great result! Have you submitted it to the leaderboard?", "replies": [ { "id": 3242129, "postDate": "2025-07-05T15:48:54.763Z", "rawMarkdown": "", "isDeleted": true } ] } ] } ], "comments": [ { "id": 3241155, "author_name": "shanzhong8", "author_url": "", "post_date": "2025-07-04T15:37:43.527000", "content": "Check this post CV-LB thread
", "votes": -2, "replies": [ { "id": 3241863, "author_name": "stefanoclss", "author_url": "", "post_date": "2025-07-05T09:39:06.300000", "content": "That's a different topic.
", "votes": 0, "replies": [] } ] }, { "id": 3243074, "author_name": "stefanoclss", "author_url": "", "post_date": "2025-07-06T18:01:00.287000", "content": "I have improved :) 0.83 cv (gbm all data) 0.75 (gbm only IMU) Public LB: 0.78
", "votes": 0, "replies": [ { "id": 3243181, "author_name": "kss", "author_url": "", "post_date": "2025-07-06T20:59:16.627000", "content": "Is that 0.78 on the public LB (and 0.75 on CV) using only IMU data, or all sensor data?
", "votes": 1, "replies": [ { "id": 3243420, "author_name": "stefanoclss", "author_url": "", "post_date": "2025-07-07T05:59:15.770000", "content": "It's both. You can check if nan's are present in your sequences.
", "votes": 0, "replies": [] } ] } ] }, { "id": 3241865, "author_name": "stefanoclss", "author_url": "", "post_date": "2025-07-05T09:40:55.727000", "content": "I wonder where I'm missing out to achieve 0.77 on IMU only data.
", "votes": 0, "replies": [] }, { "id": 3241864, "author_name": "stefanoclss", "author_url": "", "post_date": "2025-07-05T09:40:18.187000", "content": "IMU only 0.73 cv all data 0.82 cv. Public lb 0.77
", "votes": 0, "replies": [] }, { "id": 3241596, "author_name": "Fir_las_47", "author_url": "", "post_date": "2025-07-05T00:33:30.543000", "content": "is there a reason to use GBM over 1d cnn?
", "votes": 0, "replies": [ { "id": 3242128, "author_name": "kss", "author_url": "", "post_date": "2025-07-05T15:43:46.770000", "content": "No particular reason — mostly just scientific curiosity :)
", "votes": 1, "replies": [] } ] }, { "id": 3241207, "author_name": "", "author_url": "", "post_date": "2025-07-04T16:22:25.570000", "content": "", "votes": 0, "replies": [ { "id": 3242125, "author_name": "kss", "author_url": "", "post_date": "2025-07-05T15:42:56.110000", "content": "Wow, great result! Have you submitted it to the leaderboard?
", "votes": 0, "replies": [ { "id": 3242129, "author_name": "", "author_url": "", "post_date": "2025-07-05T15:48:54.763000", "content": "", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3241104": "Could you pls share what CV and leaderboard scores you achieved using handcrafted features and GBM, both with only IMU data and with all available data?\n\nI wonder how large the gap is between best gbm and neural networks on this task.\n\nHere are my scores using 5 folds StratifiedGroupKFold:\n* using IMU only. CV 0.771. Public lb 0.75", "3241155": "Check this post [CV-LB thread](https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/583023)", "3243074": "I have improved :) 0.83 cv (gbm all data) 0.75 (gbm only IMU) Public LB: 0.78", "3241865": "I wonder where I'm missing out to achieve 0.77 on IMU only data.", "3241864": "IMU only 0.73 cv all data 0.82 cv. Public lb 0.77", "3241596": "is there a reason to use GBM over 1d cnn?", "3241207": "" } }