{
  "id": 21722,
  "title": "How important is feature reduction?",
  "url": "/competitions/avito-duplicate-ads-detection/discussion/21722",
  "author_name": "",
  "post_date": "2016-06-16T10:35:08.347Z",
  "votes": null,
  "comment_count": 1,
  "views": 576,
  "content": "<p>I am a Kaggle newbie. I wonder, how important is feature reduction?</p>\n\n<p>Is anyone using clustering or PCAs or feature selection?</p>\n\n<p>I am using gradient boosting, and, other than the advantage of trees being smaller, I don't see why this should be important. Anyone knows otherwise?</p>",
  "messages": [
    {
      "id": "124223",
      "postDate": "06/16/2016 10:35:08",
      "content": "<p>I am a Kaggle newbie. I wonder, how important is feature reduction?</p>\n\n<p>Is anyone using clustering or PCAs or feature selection?</p>\n\n<p>I am using gradient boosting, and, other than the advantage of trees being smaller, I don't see why this should be important. Anyone knows otherwise?</p>",
      "rawMarkdown": "I am a Kaggle newbie. I wonder, how important is feature reduction?\r\n\r\nIs anyone using clustering or PCAs or feature selection?\r\n\r\nI am using gradient boosting, and, other than the advantage of trees being smaller, I don't see why this should be important. Anyone knows otherwise?",
      "votes": null
    },
    {
      "id": "124820",
      "postDate": "06/22/2016 15:25:10",
      "content": "<p>Gradient boosted trees are very good at ignoring useless features (xgboost, for example), so using PCA or feature selection is unlikely to help beyond speeding up the model process - and may even decrease the score due to lost information.</p>",
      "rawMarkdown": "Gradient boosted trees are very good at ignoring useless features (xgboost, for example), so using PCA or feature selection is unlikely to help beyond speeding up the model process - and may even decrease the score due to lost information.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 124820,
      "author_name": "anokas",
      "author_url": "",
      "post_date": "06/22/2016 15:25:10",
      "content": "<p>Gradient boosted trees are very good at ignoring useless features (xgboost, for example), so using PCA or feature selection is unlikely to help beyond speeding up the model process - and may even decrease the score due to lost information.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "124223": "I am a Kaggle newbie. I wonder, how important is feature reduction?\r\n\r\nIs anyone using clustering or PCAs or feature selection?\r\n\r\nI am using gradient boosting, and, other than the advantage of trees being smaller, I don't see why this should be important. Anyone knows otherwise?",
    "124820": "Gradient boosted trees are very good at ignoring useless features (xgboost, for example), so using PCA or feature selection is unlikely to help beyond speeding up the model process - and may even decrease the score due to lost information."
  },
  "source": "meta"
}