{
  "id": 542856,
  "title": "How to choose custom R2 weight if using LSTM with sequence size > 1",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/542856",
  "author_name": "",
  "post_date": "2024-10-27T08:36:49.461095900Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>As stated in the competition metric, the r2 score takes a weight for each prediction. If I'm applying LSTM to compute r2 score but with sequence size &gt; 1, which weight do I take? <br>\nThe average weight, the last timestep weight, or some other ways? </p>",
  "messages": [
    {
      "id": "3029386",
      "postDate": "10/27/2024 08:36:49",
      "content": "<p>As stated in the competition metric, the r2 score takes a weight for each prediction. If I'm applying LSTM to compute r2 score but with sequence size &gt; 1, which weight do I take? <br>\nThe average weight, the last timestep weight, or some other ways? </p>",
      "rawMarkdown": "As stated in the competition metric, the r2 score takes a weight for each prediction. If I'm applying LSTM to compute r2 score but with sequence size > 1, which weight do I take? \nThe average weight, the last timestep weight, or some other ways?",
      "votes": null
    },
    {
      "id": "3029427",
      "postDate": "10/27/2024 09:53:15",
      "content": "<p>If your lstm only outputs one value for the last time-step responder prediction, just using the weight at the last time-step should be sufficient. </p>",
      "rawMarkdown": "If your lstm only outputs one value for the last time-step responder prediction, just using the weight at the last time-step should be sufficient.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3029427,
      "author_name": "shiyili",
      "author_url": "",
      "post_date": "10/27/2024 09:53:15",
      "content": "<p>If your lstm only outputs one value for the last time-step responder prediction, just using the weight at the last time-step should be sufficient. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3029386": "As stated in the competition metric, the r2 score takes a weight for each prediction. If I'm applying LSTM to compute r2 score but with sequence size > 1, which weight do I take? \nThe average weight, the last timestep weight, or some other ways?",
    "3029427": "If your lstm only outputs one value for the last time-step responder prediction, just using the weight at the last time-step should be sufficient."
  },
  "source": "meta"
}