{
  "id": 586754,
  "title": "What's the best CV & LB using single model with IMU ONLY data",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/586754",
  "author_name": "",
  "post_date": "2025-06-28T05:44:23.919981500Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Want to understand what's the best score you can achieve with single model on IMU only data. I am exhausted on trying different models and architecture tuning, but still the best I can achieve is only at ~0.65 for a single model with IMU only data.</p>",
  "messages": [
    {
      "id": "3234621",
      "postDate": "06/28/2025 05:44:23",
      "content": "<p>Want to understand what's the best score you can achieve with single model on IMU only data. I am exhausted on trying different models and architecture tuning, but still the best I can achieve is only at ~0.65 for a single model with IMU only data.</p>",
      "rawMarkdown": "Want to understand what's the best score you can achieve with single model on IMU only data. I am exhausted on trying different models and architecture tuning, but still the best I can achieve is only at ~0.65 for a single model with IMU only data.",
      "votes": null
    },
    {
      "id": "3234623",
      "postDate": "06/28/2025 05:44:43",
      "content": "<p>want to identify if it's my model gap, or my feature engineering gap.</p>",
      "rawMarkdown": "want to identify if it's my model gap, or my feature engineering gap.",
      "votes": null
    },
    {
      "id": "3234636",
      "postDate": "06/28/2025 06:10:37",
      "content": "<p>This public notebook got 0.75 on IMU-only data <a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/kaigaokaigao/lb-0-75-imu-only-gru-inference</a></p>",
      "rawMarkdown": "This public notebook got 0.75 on IMU-only data [https://www.kaggle.com/code/kaigaokaigao/lb-0-75-imu-only-gru-inference](url)",
      "votes": null
    },
    {
      "id": "3235210",
      "postDate": "06/28/2025 21:48:35",
      "content": "<p>I think there is a dedicated discussion thread for this</p>",
      "rawMarkdown": "I think there is a dedicated discussion thread for this",
      "votes": null
    },
    {
      "id": "3235721",
      "postDate": "06/29/2025 14:59:19",
      "content": "<p>5fold cv: 0.778<br>\npublic lb: 0.78</p>\n<p>This notebook: <a href=\"https://www.kaggle.com/code/jiazhuang/cmi-imu-only-lstm\" target=\"_blank\">https://www.kaggle.com/code/jiazhuang/cmi-imu-only-lstm</a></p>",
      "rawMarkdown": "5fold cv: 0.778\npublic lb: 0.78\n\nThis notebook: https://www.kaggle.com/code/jiazhuang/cmi-imu-only-lstm",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3234623,
      "author_name": "zhangyue199",
      "author_url": "",
      "post_date": "06/28/2025 05:44:43",
      "content": "<p>want to identify if it's my model gap, or my feature engineering gap.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3234636,
      "author_name": "honganzhu",
      "author_url": "",
      "post_date": "06/28/2025 06:10:37",
      "content": "<p>This public notebook got 0.75 on IMU-only data <a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/kaigaokaigao/lb-0-75-imu-only-gru-inference</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3235210,
      "author_name": "s617603",
      "author_url": "",
      "post_date": "06/28/2025 21:48:35",
      "content": "<p>I think there is a dedicated discussion thread for this</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3235721,
      "author_name": "jiazhuang",
      "author_url": "",
      "post_date": "06/29/2025 14:59:19",
      "content": "<p>5fold cv: 0.778<br>\npublic lb: 0.78</p>\n<p>This notebook: <a href=\"https://www.kaggle.com/code/jiazhuang/cmi-imu-only-lstm\" target=\"_blank\">https://www.kaggle.com/code/jiazhuang/cmi-imu-only-lstm</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3234621": "Want to understand what's the best score you can achieve with single model on IMU only data. I am exhausted on trying different models and architecture tuning, but still the best I can achieve is only at ~0.65 for a single model with IMU only data.",
    "3234623": "want to identify if it's my model gap, or my feature engineering gap.",
    "3234636": "This public notebook got 0.75 on IMU-only data [https://www.kaggle.com/code/kaigaokaigao/lb-0-75-imu-only-gru-inference](url)",
    "3235210": "I think there is a dedicated discussion thread for this",
    "3235721": "5fold cv: 0.778\npublic lb: 0.78\n\nThis notebook: https://www.kaggle.com/code/jiazhuang/cmi-imu-only-lstm"
  },
  "source": "meta"
}