{
  "id": 200977,
  "title": "Using lectures for sequence models?",
  "url": "/competitions/riiid-test-answer-prediction/discussion/200977",
  "author_name": "Hamish Scott",
  "post_date": "2020-12-02T15:46:26.642000",
  "votes": 4,
  "comment_count": 0,
  "views": 0,
  "content": "<p>How useful have people found lectures? </p>\n<p>It seems using lectures for feature engineering can help with tabular approaches e.g. gbm but what about sequence models? </p>\n<p>I have tried including lectures in a sequence model but seen essentially no improvement in local validation and none at all in lb. This might be because of the low relative frequency of lectures in the data.</p>\n<p>What have others found?</p>",
  "messages": [
    {
      "id": 1099796,
      "postDate": "2020-12-02T15:46:26.643Z",
      "content": "<p>How useful have people found lectures? </p>\n<p>It seems using lectures for feature engineering can help with tabular approaches e.g. gbm but what about sequence models? </p>\n<p>I have tried including lectures in a sequence model but seen essentially no improvement in local validation and none at all in lb. This might be because of the low relative frequency of lectures in the data.</p>\n<p>What have others found?</p>",
      "rawMarkdown": "How useful have people found lectures? \n\nIt seems using lectures for feature engineering can help with tabular approaches e.g. gbm but what about sequence models? \n\nI have tried including lectures in a sequence model but seen essentially no improvement in local validation and none at all in lb. This might be because of the low relative frequency of lectures in the data.\n\nWhat have others found?",
      "votes": 4
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1099796": "How useful have people found lectures? \n\nIt seems using lectures for feature engineering can help with tabular approaches e.g. gbm but what about sequence models? \n\nI have tried including lectures in a sequence model but seen essentially no improvement in local validation and none at all in lb. This might be because of the low relative frequency of lectures in the data.\n\nWhat have others found?"
  }
}