{
  "id": 68182,
  "title": "Ensembling methods",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/68182",
  "author_name": "",
  "post_date": "2018-10-10T07:06:37.941994500Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>It seems like most of top score public models uses weighted mean value of models predictions to make submission file. Does anybody tried different ways to combine models (e.g. harmonic or geometric mean)?</p>",
  "messages": [
    {
      "id": "401496",
      "postDate": "10/10/2018 07:06:37",
      "content": "<p>It seems like most of top score public models uses weighted mean value of models predictions to make submission file. Does anybody tried different ways to combine models (e.g. harmonic or geometric mean)?</p>",
      "rawMarkdown": "It seems like most of top score public models uses weighted mean value of models predictions to make submission file. Does anybody tried different ways to combine models (e.g. harmonic or geometric mean)?",
      "votes": null
    },
    {
      "id": "401870",
      "postDate": "10/10/2018 20:08:22",
      "content": "<p>I tried, and the results were horrible (CV to 1.8 from ~1.54)</p>",
      "rawMarkdown": "I tried, and the results were horrible (CV to 1.8 from ~1.54)",
      "votes": null
    },
    {
      "id": "402625",
      "postDate": "10/12/2018 02:41:30",
      "content": "<p>I don't know how you combined models.</p>\n\n<p>But If your models themselves have good scores and there is lower correlation between Them - you can get a better score(most likely)</p>",
      "rawMarkdown": "I don't know how you combined models.\n\nBut If your models themselves have good scores and there is lower correlation between Them - you can get a better score(most likely)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 401870,
      "author_name": "dkaraflos",
      "author_url": "",
      "post_date": "10/10/2018 20:08:22",
      "content": "<p>I tried, and the results were horrible (CV to 1.8 from ~1.54)</p>",
      "votes": null,
      "replies": [
        {
          "id": 402625,
          "author_name": "hemantrupani",
          "author_url": "",
          "post_date": "10/12/2018 02:41:30",
          "content": "<p>I don't know how you combined models.</p>\n\n<p>But If your models themselves have good scores and there is lower correlation between Them - you can get a better score(most likely)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "401496": "It seems like most of top score public models uses weighted mean value of models predictions to make submission file. Does anybody tried different ways to combine models (e.g. harmonic or geometric mean)?",
    "401870": "I tried, and the results were horrible (CV to 1.8 from ~1.54)",
    "402625": "I don't know how you combined models.\n\nBut If your models themselves have good scores and there is lower correlation between Them - you can get a better score(most likely)"
  },
  "source": "meta"
}