{
  "id": 44874,
  "title": "Using NWRMSLE as eval_metric on XGBoost?",
  "url": "/competitions/favorita-grocery-sales-forecasting/discussion/44874",
  "author_name": "",
  "post_date": "2017-12-03T18:04:15.253177900Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi all,</p>\n\n<p>This is my first time participating in a Kaggle competition so excuse this question if it's a bit stupid.</p>\n\n<p>I saw one kernel doing XGBoost and they used some sort of rmse for the eval_metric... Doesn't it make sense to set the eval_metric as the NWRMSLE? Or are there problems with doing that such as overfitting?</p>",
  "messages": [
    {
      "id": "252731",
      "postDate": "12/03/2017 18:04:15",
      "content": "<p>Hi all,</p>\n\n<p>This is my first time participating in a Kaggle competition so excuse this question if it's a bit stupid.</p>\n\n<p>I saw one kernel doing XGBoost and they used some sort of rmse for the eval_metric... Doesn't it make sense to set the eval_metric as the NWRMSLE? Or are there problems with doing that such as overfitting?</p>",
      "rawMarkdown": "Hi all,\n\nThis is my first time participating in a Kaggle competition so excuse this question if it's a bit stupid.\n\nI saw one kernel doing XGBoost and they used some sort of rmse for the eval_metric... Doesn't it make sense to set the eval_metric as the NWRMSLE? Or are there problems with doing that such as overfitting?",
      "votes": null
    },
    {
      "id": "252766",
      "postDate": "12/03/2017 20:09:22",
      "content": "<p>I do the same RMSE training loss because I'm feeding my model LN(sales+1) as my main method of normalization.  Perhaps that's what they're doing as well. </p>",
      "rawMarkdown": "I do the same RMSE training loss because I'm feeding my model LN(sales+1) as my main method of normalization.  Perhaps that's what they're doing as well.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 252766,
      "author_name": "vannak",
      "author_url": "",
      "post_date": "12/03/2017 20:09:22",
      "content": "<p>I do the same RMSE training loss because I'm feeding my model LN(sales+1) as my main method of normalization.  Perhaps that's what they're doing as well. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "252731": "Hi all,\n\nThis is my first time participating in a Kaggle competition so excuse this question if it's a bit stupid.\n\nI saw one kernel doing XGBoost and they used some sort of rmse for the eval_metric... Doesn't it make sense to set the eval_metric as the NWRMSLE? Or are there problems with doing that such as overfitting?",
    "252766": "I do the same RMSE training loss because I'm feeding my model LN(sales+1) as my main method of normalization.  Perhaps that's what they're doing as well."
  },
  "source": "meta"
}