{
  "id": 550679,
  "title": "Is it feasible to employ LightGBM's boost.update for online learning purposes?",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/550679",
  "author_name": "JWQ",
  "post_date": "2024-12-09T00:55:06.798000",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I am curious as to whether there would be an enhancement when leveraging the daily lag data to update the pre-trained model. </p>\n<p>However, I constantly encounter the following error, even though I have already set keep_training_booster=True as per <a href=\"https://github.com/microsoft/LightGBM/issues/4466\" target=\"_blank\">this link</a>. </p>\n<p>The error message is:<br>\n<code>'Booster' object has no attribute 'train_set'</code></p>\n<p>Any assistance would be greatly appreciated.</p>",
  "messages": [
    {
      "id": 3067143,
      "postDate": "2024-12-09T00:55:06.800Z",
      "content": "<p>I am curious as to whether there would be an enhancement when leveraging the daily lag data to update the pre-trained model. </p>\n<p>However, I constantly encounter the following error, even though I have already set keep_training_booster=True as per <a href=\"https://github.com/microsoft/LightGBM/issues/4466\" target=\"_blank\">this link</a>. </p>\n<p>The error message is:<br>\n<code>'Booster' object has no attribute 'train_set'</code></p>\n<p>Any assistance would be greatly appreciated.</p>",
      "rawMarkdown": "I am curious as to whether there would be an enhancement when leveraging the daily lag data to update the pre-trained model. \n\nHowever, I constantly encounter the following error, even though I have already set keep_training_booster=True as per [this link](https://github.com/microsoft/LightGBM/issues/4466). \n\nThe error message is:\n`'Booster' object has no attribute 'train_set'`\n\nAny assistance would be greatly appreciated.",
      "votes": 2
    },
    {
      "id": 3067287,
      "postDate": "2024-12-09T06:15:16.763Z",
      "content": "<p>Yes, this is possible, but one will need to check if this will work within the time constraint of 1 min per batch served in the test dataset <a href=\"https://www.kaggle.com/carrotwait\" target=\"_blank\">@carrotwait</a> </p>\n<p>In my opinion, one will not be able to use online learning with GBM models here due to time and data size constraints.</p>",
      "rawMarkdown": "Yes, this is possible, but one will need to check if this will work within the time constraint of 1 min per batch served in the test dataset @carrotwait \n\nIn my opinion, one will not be able to use online learning with GBM models here due to time and data size constraints.",
      "replies": [
        {
          "id": 3067476,
          "postDate": "2024-12-09T10:24:08.687Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 3067146,
      "postDate": "2024-12-09T01:08:29.890Z",
      "content": "<p>i think you need convert the data into lgbm class</p>",
      "rawMarkdown": "i think you need convert the data into lgbm class"
    }
  ],
  "comments": [
    {
      "id": 3067287,
      "author_name": "Ravi Ramakrishnan",
      "author_url": "",
      "post_date": "2024-12-09T06:15:16.763000",
      "content": "<p>Yes, this is possible, but one will need to check if this will work within the time constraint of 1 min per batch served in the test dataset <a href=\"https://www.kaggle.com/carrotwait\" target=\"_blank\">@carrotwait</a> </p>\n<p>In my opinion, one will not be able to use online learning with GBM models here due to time and data size constraints.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 3067476,
          "author_name": "",
          "author_url": "",
          "post_date": "2024-12-09T10:24:08.687000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 3067146,
      "author_name": "ZT",
      "author_url": "",
      "post_date": "2024-12-09T01:08:29.890000",
      "content": "<p>i think you need convert the data into lgbm class</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3067143": "I am curious as to whether there would be an enhancement when leveraging the daily lag data to update the pre-trained model. \n\nHowever, I constantly encounter the following error, even though I have already set keep_training_booster=True as per [this link](https://github.com/microsoft/LightGBM/issues/4466). \n\nThe error message is:\n`'Booster' object has no attribute 'train_set'`\n\nAny assistance would be greatly appreciated.",
    "3067287": "Yes, this is possible, but one will need to check if this will work within the time constraint of 1 min per batch served in the test dataset @carrotwait \n\nIn my opinion, one will not be able to use online learning with GBM models here due to time and data size constraints.",
    "3067146": "i think you need convert the data into lgbm class"
  }
}