{
  "id": 585513,
  "title": "Past Kaggle Contests",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/585513",
  "author_name": "Noah Stefancik",
  "post_date": "2025-06-20T23:23:32.839000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Currently I am still working with only the IMU data as I want to create a great model for this before moving on. I started with gradient boosting methods such as lgbm, xgboost, and catboost which I am comfortable with and was able to obtain a score of 0.69. From there I have moved onto network models such as 1dCNN, LSTM, GRU, experimenting with attention and all sort of things and have achieved a CV score of 0.74. I'm fairly new to such models and have read papers and such to get myself here. Does anyone have any similar past competitions or implantations of such models that I can learn from to better design a model. or any tips for model architecture. I have found the best success with the 1dCNN so far. Just curious as I want to learn these techniques the correct way rather than sloppily put together model that does well on the LB.</p>",
  "messages": [
    {
      "id": 3229053,
      "postDate": "2025-06-20T23:23:32.840Z",
      "content": "<p>Currently I am still working with only the IMU data as I want to create a great model for this before moving on. I started with gradient boosting methods such as lgbm, xgboost, and catboost which I am comfortable with and was able to obtain a score of 0.69. From there I have moved onto network models such as 1dCNN, LSTM, GRU, experimenting with attention and all sort of things and have achieved a CV score of 0.74. I'm fairly new to such models and have read papers and such to get myself here. Does anyone have any similar past competitions or implantations of such models that I can learn from to better design a model. or any tips for model architecture. I have found the best success with the 1dCNN so far. Just curious as I want to learn these techniques the correct way rather than sloppily put together model that does well on the LB.</p>",
      "rawMarkdown": "Currently I am still working with only the IMU data as I want to create a great model for this before moving on. I started with gradient boosting methods such as lgbm, xgboost, and catboost which I am comfortable with and was able to obtain a score of 0.69. From there I have moved onto network models such as 1dCNN, LSTM, GRU, experimenting with attention and all sort of things and have achieved a CV score of 0.74. I'm fairly new to such models and have read papers and such to get myself here. Does anyone have any similar past competitions or implantations of such models that I can learn from to better design a model. or any tips for model architecture. I have found the best success with the 1dCNN so far. Just curious as I want to learn these techniques the correct way rather than sloppily put together model that does well on the LB.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3229053": "Currently I am still working with only the IMU data as I want to create a great model for this before moving on. I started with gradient boosting methods such as lgbm, xgboost, and catboost which I am comfortable with and was able to obtain a score of 0.69. From there I have moved onto network models such as 1dCNN, LSTM, GRU, experimenting with attention and all sort of things and have achieved a CV score of 0.74. I'm fairly new to such models and have read papers and such to get myself here. Does anyone have any similar past competitions or implantations of such models that I can learn from to better design a model. or any tips for model architecture. I have found the best success with the 1dCNN so far. Just curious as I want to learn these techniques the correct way rather than sloppily put together model that does well on the LB."
  }
}