{
  "id": 116162,
  "title": "Why most of the winning solution choose regression approach over classification?",
  "url": "/competitions/aptos2019-blindness-detection/discussion/116162",
  "author_name": "",
  "post_date": "2019-11-07T11:01:05.914449Z",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "",
  "messages": [
    {
      "id": "667544",
      "postDate": "11/07/2019 11:01:05",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "674378",
      "postDate": "11/16/2019 11:18:40",
      "content": "<p>Because of this competition uses Quadratic Weighted Kappa as evaluation metric. It's highly related to \"Mean squared error\" loss function. So by using regression approach local CV is highly correlated to the final private LB (Although not Public LB due to QWK is highly dependent to classes distribution).</p>",
      "rawMarkdown": "Because of this competition uses Quadratic Weighted Kappa as evaluation metric. It's highly related to \"Mean squared error\" loss function. So by using regression approach local CV is highly correlated to the final private LB (Although not Public LB due to QWK is highly dependent to classes distribution).",
      "votes": null
    },
    {
      "id": "2474433",
      "postDate": "10/09/2023 08:12:23",
      "content": "<p>Thank you for this insight (questionnaire, answerer)</p>",
      "rawMarkdown": "Thank you for this insight (questionnaire, answerer)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 674378,
      "author_name": "quandapro",
      "author_url": "",
      "post_date": "11/16/2019 11:18:40",
      "content": "<p>Because of this competition uses Quadratic Weighted Kappa as evaluation metric. It's highly related to \"Mean squared error\" loss function. So by using regression approach local CV is highly correlated to the final private LB (Although not Public LB due to QWK is highly dependent to classes distribution).</p>",
      "votes": null,
      "replies": [
        {
          "id": 2474433,
          "author_name": "sarvagyaporwal",
          "author_url": "",
          "post_date": "10/09/2023 08:12:23",
          "content": "<p>Thank you for this insight (questionnaire, answerer)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "667544": "",
    "674378": "Because of this competition uses Quadratic Weighted Kappa as evaluation metric. It's highly related to \"Mean squared error\" loss function. So by using regression approach local CV is highly correlated to the final private LB (Although not Public LB due to QWK is highly dependent to classes distribution).",
    "2474433": "Thank you for this insight (questionnaire, answerer)"
  },
  "source": "meta"
}