{
  "id": 114458,
  "title": "Any sample weighting strategies?",
  "url": "/competitions/ashrae-energy-prediction/discussion/114458",
  "author_name": "",
  "post_date": "2019-10-26T12:11:58.841016900Z",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi guys!\n<br>Eventhough <code>rmsle</code> metric is quite robust to higher absolute deviations, say <code>rmsle(5, 10) ~ rmsle(5000, 10000)</code> I wonder whether sample weighting strategies, from simple <br><code>weights = (target + lambda) / (lambda + target.max())</code> to more sophisticated ones works here.\n<br>What do you think? Did anyone had the success with weighting?</p>",
  "messages": [
    {
      "id": "658711",
      "postDate": "10/26/2019 12:11:58",
      "content": "<p>Hi guys!\n<br>Eventhough <code>rmsle</code> metric is quite robust to higher absolute deviations, say <code>rmsle(5, 10) ~ rmsle(5000, 10000)</code> I wonder whether sample weighting strategies, from simple <br><code>weights = (target + lambda) / (lambda + target.max())</code> to more sophisticated ones works here.\n<br>What do you think? Did anyone had the success with weighting?</p>",
      "rawMarkdown": "Hi guys!\n<br>Eventhough `rmsle` metric is quite robust to higher absolute deviations, say `rmsle(5, 10) ~ rmsle(5000, 10000)` I wonder whether sample weighting strategies, from simple <br>`weights = (target + lambda) / (lambda + target.max())` to more sophisticated ones works here.\n<br>What do you think? Did anyone had the success with weighting?",
      "votes": null
    },
    {
      "id": "678912",
      "postDate": "11/22/2019 03:31:42",
      "content": "<p>Hi, Did you use this strategy in the end ? Did it help you ? </p>",
      "rawMarkdown": "Hi, Did you use this strategy in the end ? Did it help you ?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 678912,
      "author_name": "alicegabriel",
      "author_url": "",
      "post_date": "11/22/2019 03:31:42",
      "content": "<p>Hi, Did you use this strategy in the end ? Did it help you ? </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "658711": "Hi guys!\n<br>Eventhough `rmsle` metric is quite robust to higher absolute deviations, say `rmsle(5, 10) ~ rmsle(5000, 10000)` I wonder whether sample weighting strategies, from simple <br>`weights = (target + lambda) / (lambda + target.max())` to more sophisticated ones works here.\n<br>What do you think? Did anyone had the success with weighting?",
    "678912": "Hi, Did you use this strategy in the end ? Did it help you ?"
  },
  "source": "meta"
}