{
  "id": 115246,
  "title": "One simple way to let your model know local time.",
  "url": "/competitions/ashrae-energy-prediction/discussion/115246",
  "author_name": "",
  "post_date": "2019-11-01T09:12:13.438728100Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Combining site and day together  and do one-hot encoding. </p>\n\n<ul>\n<li>OHE(df['site'] + '_' + df['day']) </li>\n</ul>\n\n<p>Since Tree based models split one feature into two parts each time. This helps your model know local time well. But it will lose some statistical infos and needs more memory. </p>",
  "messages": [
    {
      "id": "662961",
      "postDate": "11/01/2019 09:12:13",
      "content": "<p>Combining site and day together  and do one-hot encoding. </p>\n\n<ul>\n<li>OHE(df['site'] + '_' + df['day']) </li>\n</ul>\n\n<p>Since Tree based models split one feature into two parts each time. This helps your model know local time well. But it will lose some statistical infos and needs more memory. </p>",
      "rawMarkdown": "Combining site and day together  and do one-hot encoding. \n\n- OHE(df['site'] + '_' + df['day']) \n\nSince Tree based models split one feature into two parts each time. This helps your model know local time well. But it will lose some statistical infos and needs more memory.",
      "votes": null
    },
    {
      "id": "662988",
      "postDate": "11/01/2019 09:41:26",
      "content": "<p>Thanks for sharing but this will generate too many features. I think choosing a location and standardizing the time relatively to that location is a better approach.</p>",
      "rawMarkdown": "Thanks for sharing but this will generate too many features. I think choosing a location and standardizing the time relatively to that location is a better approach.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 662988,
      "author_name": "gunesevitan",
      "author_url": "",
      "post_date": "11/01/2019 09:41:26",
      "content": "<p>Thanks for sharing but this will generate too many features. I think choosing a location and standardizing the time relatively to that location is a better approach.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "662961": "Combining site and day together  and do one-hot encoding. \n\n- OHE(df['site'] + '_' + df['day']) \n\nSince Tree based models split one feature into two parts each time. This helps your model know local time well. But it will lose some statistical infos and needs more memory.",
    "662988": "Thanks for sharing but this will generate too many features. I think choosing a location and standardizing the time relatively to that location is a better approach."
  },
  "source": "meta"
}