{
  "id": 205338,
  "title": "While plotting, the feature importance of categorical feature is extremely high.",
  "url": "/competitions/riiid-test-answer-prediction/discussion/205338",
  "author_name": "Shihao Shao",
  "post_date": "2020-12-19T16:44:57.393000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I have trained LGB model without categorical feature before. However, I recently added 'bundle_id' feature as a categorical feature. The werid thing (at least IMO) is that, when I tried to plot the feature importance, the importance of 'bundle_id' reached 39,000+ which is about 15x higher than feature in the second place. </p>\n<p>Is there something wrong or it just should be like that?</p>",
  "messages": [
    {
      "id": 1119030,
      "postDate": "2020-12-19T16:44:57.393Z",
      "content": "<p>I have trained LGB model without categorical feature before. However, I recently added 'bundle_id' feature as a categorical feature. The werid thing (at least IMO) is that, when I tried to plot the feature importance, the importance of 'bundle_id' reached 39,000+ which is about 15x higher than feature in the second place. </p>\n<p>Is there something wrong or it just should be like that?</p>",
      "rawMarkdown": "I have trained LGB model without categorical feature before. However, I recently added 'bundle_id' feature as a categorical feature. The werid thing (at least IMO) is that, when I tried to plot the feature importance, the importance of 'bundle_id' reached 39,000+ which is about 15x higher than feature in the second place. \n\nIs there something wrong or it just should be like that?"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1119030": "I have trained LGB model without categorical feature before. However, I recently added 'bundle_id' feature as a categorical feature. The werid thing (at least IMO) is that, when I tried to plot the feature importance, the importance of 'bundle_id' reached 39,000+ which is about 15x higher than feature in the second place. \n\nIs there something wrong or it just should be like that?"
  }
}