{
  "id": 117652,
  "title": "Cross-validation technique ",
  "url": "/competitions/ashrae-energy-prediction/discussion/117652",
  "author_name": "",
  "post_date": "2019-11-17T00:03:26.388300300Z",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi!\nWhich cross-validation strategy do you think is best for this competition?\nMaybe it's better to use the building_id or smth else instead of time? What do you think?</p>",
  "messages": [
    {
      "id": "674711",
      "postDate": "11/17/2019 00:03:26",
      "content": "<p>Hi!\nWhich cross-validation strategy do you think is best for this competition?\nMaybe it's better to use the building_id or smth else instead of time? What do you think?</p>",
      "rawMarkdown": "Hi!\nWhich cross-validation strategy do you think is best for this competition?\nMaybe it's better to use the building_id or smth else instead of time? What do you think?",
      "votes": null
    },
    {
      "id": "676455",
      "postDate": "11/19/2019 07:41:33",
      "content": "<p>Using the building_id will almost definitely cause you to estimate worse performance than you'll get on the LB. This competition is all about learning the characteristics of each building, so forcing your CV to predict for unknown buildings won't be useful. I've had pretty good predictive accuracy with a time-based split. </p>",
      "rawMarkdown": "Using the building_id will almost definitely cause you to estimate worse performance than you'll get on the LB. This competition is all about learning the characteristics of each building, so forcing your CV to predict for unknown buildings won't be useful. I've had pretty good predictive accuracy with a time-based split.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 676455,
      "author_name": "purist1024",
      "author_url": "",
      "post_date": "11/19/2019 07:41:33",
      "content": "<p>Using the building_id will almost definitely cause you to estimate worse performance than you'll get on the LB. This competition is all about learning the characteristics of each building, so forcing your CV to predict for unknown buildings won't be useful. I've had pretty good predictive accuracy with a time-based split. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "674711": "Hi!\nWhich cross-validation strategy do you think is best for this competition?\nMaybe it's better to use the building_id or smth else instead of time? What do you think?",
    "676455": "Using the building_id will almost definitely cause you to estimate worse performance than you'll get on the LB. This competition is all about learning the characteristics of each building, so forcing your CV to predict for unknown buildings won't be useful. I've had pretty good predictive accuracy with a time-based split."
  },
  "source": "meta"
}