{
  "id": 555884,
  "title": "Per-sample weights for online retraining",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/555884",
  "author_name": "",
  "post_date": "2025-01-09T18:50:43.671708800Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Are per-sample weights available for online retraining? During initial training, loss was weighted by the per-sample weights. Does this hurt the performance of NN during online retraining?</p>\n<p>I would appreciate it if someone could help clear up my confusion.</p>",
  "messages": [
    {
      "id": "3092608",
      "postDate": "01/09/2025 18:50:43",
      "content": "<p>Are per-sample weights available for online retraining? During initial training, loss was weighted by the per-sample weights. Does this hurt the performance of NN during online retraining?</p>\n<p>I would appreciate it if someone could help clear up my confusion.</p>",
      "rawMarkdown": "Are per-sample weights available for online retraining? During initial training, loss was weighted by the per-sample weights. Does this hurt the performance of NN during online retraining?\n\nI would appreciate it if someone could help clear up my confusion.",
      "votes": null
    },
    {
      "id": "3092670",
      "postDate": "01/09/2025 20:31:04",
      "content": "<blockquote>\n  <p>Are per-sample weights available for online retraining?</p>\n</blockquote>\n<p>The test data set we receive (for the first time via the predict() call) has a weight column.  If you need to use these test data and their weights at a later time for online learning purposes, you can always save them for later usage.</p>\n<p>I am assuming that the weight of any data point stays fixed once we receive the data set for the first time.  Someone please correct me if I'm wrong.</p>",
      "rawMarkdown": ">Are per-sample weights available for online retraining?\n\nThe test data set we receive (for the first time via the predict() call) has a weight column.  If you need to use these test data and their weights at a later time for online learning purposes, you can always save them for later usage.\n\nI am assuming that the weight of any data point stays fixed once we receive the data set for the first time.  Someone please correct me if I'm wrong.",
      "votes": null
    },
    {
      "id": "3092740",
      "postDate": "01/10/2025 00:25:10",
      "content": "<p>I see. Thanks for the response</p>",
      "rawMarkdown": "I see. Thanks for the response",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3092670,
      "author_name": "revealer",
      "author_url": "",
      "post_date": "01/09/2025 20:31:04",
      "content": "<blockquote>\n  <p>Are per-sample weights available for online retraining?</p>\n</blockquote>\n<p>The test data set we receive (for the first time via the predict() call) has a weight column.  If you need to use these test data and their weights at a later time for online learning purposes, you can always save them for later usage.</p>\n<p>I am assuming that the weight of any data point stays fixed once we receive the data set for the first time.  Someone please correct me if I'm wrong.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3092740,
          "author_name": "kaleab7",
          "author_url": "",
          "post_date": "01/10/2025 00:25:10",
          "content": "<p>I see. Thanks for the response</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3092608": "Are per-sample weights available for online retraining? During initial training, loss was weighted by the per-sample weights. Does this hurt the performance of NN during online retraining?\n\nI would appreciate it if someone could help clear up my confusion.",
    "3092670": ">Are per-sample weights available for online retraining?\n\nThe test data set we receive (for the first time via the predict() call) has a weight column.  If you need to use these test data and their weights at a later time for online learning purposes, you can always save them for later usage.\n\nI am assuming that the weight of any data point stays fixed once we receive the data set for the first time.  Someone please correct me if I'm wrong.",
    "3092740": "I see. Thanks for the response"
  },
  "source": "meta"
}