{
  "id": 455861,
  "title": "transformed anglez feature",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/455861",
  "author_name": "",
  "post_date": "2023-11-16T18:47:18.468925400Z",
  "votes": 7,
  "comment_count": 3,
  "views": 0,
  "content": "<p>personally i cant be competitive in this competition because the resource requirements (especially time to train on gpu) are way too large for kaggle notebooks to handle and i dont wanna spend money. so i will just share a feature which ive found to be much better than both anglez or enmo, maybe it helps someone out a bit.</p>\n<p><code>abs(df.groupby('series_id')['anglez'].diff())</code></p>\n<p>this feature is visually much more obvious to tell when someone falls asleep and personally for my models ive found them to like this feature more than anglez and enmo</p>",
  "messages": [
    {
      "id": "2527743",
      "postDate": "11/16/2023 18:47:18",
      "content": "<p>personally i cant be competitive in this competition because the resource requirements (especially time to train on gpu) are way too large for kaggle notebooks to handle and i dont wanna spend money. so i will just share a feature which ive found to be much better than both anglez or enmo, maybe it helps someone out a bit.</p>\n<p><code>abs(df.groupby('series_id')['anglez'].diff())</code></p>\n<p>this feature is visually much more obvious to tell when someone falls asleep and personally for my models ive found them to like this feature more than anglez and enmo</p>",
      "rawMarkdown": "personally i cant be competitive in this competition because the resource requirements (especially time to train on gpu) are way too large for kaggle notebooks to handle and i dont wanna spend money. so i will just share a feature which ive found to be much better than both anglez or enmo, maybe it helps someone out a bit.\n\n`abs(df.groupby('series_id')['anglez'].diff())`\n\nthis feature is visually much more obvious to tell when someone falls asleep and personally for my models ive found them to like this feature more than anglez and enmo",
      "votes": null
    },
    {
      "id": "2529056",
      "postDate": "11/17/2023 21:48:07",
      "content": "<p>Thank you for sharing this feature. Actually you may use free Google Colab to train models.<br>\nPlus, which model are you using? A tree-based model like XGBoost or a Transformer architecture?</p>",
      "rawMarkdown": "Thank you for sharing this feature. Actually you may use free Google Colab to train models.\nPlus, which model are you using? A tree-based model like XGBoost or a Transformer architecture?",
      "votes": null
    },
    {
      "id": "2529768",
      "postDate": "11/18/2023 14:26:17",
      "content": "<p>Thanks for points out, i'll give it a try, but could you explain why this feature can be more useful than others ? <a href=\"https://www.kaggle.com/at7459\" target=\"_blank\">@at7459</a> </p>",
      "rawMarkdown": "Thanks for points out, i'll give it a try, but could you explain why this feature can be more useful than others ? @at7459",
      "votes": null
    },
    {
      "id": "2529940",
      "postDate": "11/18/2023 17:02:51",
      "content": "<p>Impressive job.Thanks for your sharing.I will try to implement it.</p>",
      "rawMarkdown": "Impressive job.Thanks for your sharing.I will try to implement it.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2529056,
      "author_name": "lizhecheng",
      "author_url": "",
      "post_date": "11/17/2023 21:48:07",
      "content": "<p>Thank you for sharing this feature. Actually you may use free Google Colab to train models.<br>\nPlus, which model are you using? A tree-based model like XGBoost or a Transformer architecture?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2529768,
      "author_name": "liushundawang",
      "author_url": "",
      "post_date": "11/18/2023 14:26:17",
      "content": "<p>Thanks for points out, i'll give it a try, but could you explain why this feature can be more useful than others ? <a href=\"https://www.kaggle.com/at7459\" target=\"_blank\">@at7459</a> </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2529940,
      "author_name": "hugginglook",
      "author_url": "",
      "post_date": "11/18/2023 17:02:51",
      "content": "<p>Impressive job.Thanks for your sharing.I will try to implement it.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2527743": "personally i cant be competitive in this competition because the resource requirements (especially time to train on gpu) are way too large for kaggle notebooks to handle and i dont wanna spend money. so i will just share a feature which ive found to be much better than both anglez or enmo, maybe it helps someone out a bit.\n\n`abs(df.groupby('series_id')['anglez'].diff())`\n\nthis feature is visually much more obvious to tell when someone falls asleep and personally for my models ive found them to like this feature more than anglez and enmo",
    "2529056": "Thank you for sharing this feature. Actually you may use free Google Colab to train models.\nPlus, which model are you using? A tree-based model like XGBoost or a Transformer architecture?",
    "2529768": "Thanks for points out, i'll give it a try, but could you explain why this feature can be more useful than others ? @at7459",
    "2529940": "Impressive job.Thanks for your sharing.I will try to implement it."
  },
  "source": "meta"
}