{
  "id": 70422,
  "title": "Hits-Skewness - Box Cox transformation",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/70422",
  "author_name": "",
  "post_date": "2018-11-03T12:47:43.105315900Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi All,\nTried to handle Hits' skewness  with Box Cox. It did not improve results with randomforest. \nI guess BCox works better with linear regression models than tree models.</p>",
  "messages": [
    {
      "id": "414713",
      "postDate": "11/03/2018 12:47:43",
      "content": "<p>Hi All,\nTried to handle Hits' skewness  with Box Cox. It did not improve results with randomforest. \nI guess BCox works better with linear regression models than tree models.</p>",
      "rawMarkdown": "Hi All,\nTried to handle Hits' skewness  with Box Cox. It did not improve results with randomforest. \nI guess BCox works better with linear regression models than tree models.",
      "votes": null
    },
    {
      "id": "415846",
      "postDate": "11/05/2018 19:42:32",
      "content": "<p>I don’t think monotonic transforms of predictors make any difference to decision tree models. Unless it affects some kind of optimisation, like feature value bucketing in xgboost.</p>",
      "rawMarkdown": "I don’t think monotonic transforms of predictors make any difference to decision tree models. Unless it affects some kind of optimisation, like feature value bucketing in xgboost.",
      "votes": null
    },
    {
      "id": "426372",
      "postDate": "11/23/2018 06:39:44",
      "content": "<p>Why you need normality here? It is designed for inference in linear models not prediction. And it is usually applied to response not predictors.</p>",
      "rawMarkdown": "Why you need normality here? It is designed for inference in linear models not prediction. And it is usually applied to response not predictors.",
      "votes": null
    },
    {
      "id": "427360",
      "postDate": "11/25/2018 10:22:10",
      "content": "<p>Thank you ZChen</p>",
      "rawMarkdown": "Thank you ZChen",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 415846,
      "author_name": "burritodan",
      "author_url": "",
      "post_date": "11/05/2018 19:42:32",
      "content": "<p>I don’t think monotonic transforms of predictors make any difference to decision tree models. Unless it affects some kind of optimisation, like feature value bucketing in xgboost.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 426372,
      "author_name": "czhfrank",
      "author_url": "",
      "post_date": "11/23/2018 06:39:44",
      "content": "<p>Why you need normality here? It is designed for inference in linear models not prediction. And it is usually applied to response not predictors.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 427360,
      "author_name": "antonspartan",
      "author_url": "",
      "post_date": "11/25/2018 10:22:10",
      "content": "<p>Thank you ZChen</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "414713": "Hi All,\nTried to handle Hits' skewness  with Box Cox. It did not improve results with randomforest. \nI guess BCox works better with linear regression models than tree models.",
    "415846": "I don’t think monotonic transforms of predictors make any difference to decision tree models. Unless it affects some kind of optimisation, like feature value bucketing in xgboost.",
    "426372": "Why you need normality here? It is designed for inference in linear models not prediction. And it is usually applied to response not predictors.",
    "427360": "Thank you ZChen"
  },
  "source": "meta"
}