{
  "id": 345905,
  "title": "Evaluation Function",
  "url": "/competitions/mayo-clinic-strip-ai/discussion/345905",
  "author_name": "Raghav Garg 12",
  "post_date": "2022-08-17T03:59:43.204000",
  "votes": 10,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Submissions are evaluated using a weighted multi-class logarithmic loss. This function is given under Overview-&gt;Evaluation.<br>\nAll the parameters have been defined except 'w'. I thought it to be weights but it doesn't satisfy the equation mathematically.. Any idea what 'w' might be ?</p>",
  "messages": [
    {
      "id": 2682440,
      "postDate": "2024-03-05T10:27:52.457Z",
      "content": "<p>But how will it work with linear regression</p>",
      "rawMarkdown": "But how will it work with linear regression",
      "votes": 1
    },
    {
      "id": 1903016,
      "postDate": "2022-08-17T03:59:43.203Z",
      "content": "<p>Submissions are evaluated using a weighted multi-class logarithmic loss. This function is given under Overview-&gt;Evaluation.<br>\nAll the parameters have been defined except 'w'. I thought it to be weights but it doesn't satisfy the equation mathematically.. Any idea what 'w' might be ?</p>",
      "rawMarkdown": "Submissions are evaluated using a weighted multi-class logarithmic loss. This function is given under Overview->Evaluation.\nAll the parameters have been defined except 'w'. I thought it to be weights but it doesn't satisfy the equation mathematically.. Any idea what 'w' might be ?",
      "votes": 10
    },
    {
      "id": 1909842,
      "postDate": "2022-08-23T01:33:00.127Z",
      "content": "<p>My guess would be that weights are kept a secret to avoid probing from strategic submissions, so that the public score provides a good indication but also doesn't give away any leak from test data. Most likely we have to rely on our own log_loss performance on a hold out set. But I could be wrong :)</p>",
      "rawMarkdown": "My guess would be that weights are kept a secret to avoid probing from strategic submissions, so that the public score provides a good indication but also doesn't give away any leak from test data. Most likely we have to rely on our own log_loss performance on a hold out set. But I could be wrong :)",
      "votes": 1
    },
    {
      "id": 1904670,
      "postDate": "2022-08-18T11:42:42.273Z",
      "content": "<p>I understood 'wi' to be weight for each class.<br>\nBut it doesn't really specify what weights are given to each of the 2 classes - CE and LAA here<br>\nMaybe they are both given 0.5 ?<br>\nClarification on this would be good</p>",
      "rawMarkdown": "I understood 'wi' to be weight for each class.\nBut it doesn't really specify what weights are given to each of the 2 classes - CE and LAA here\nMaybe they are both given 0.5 ?\nClarification on this would be good",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 2682440,
      "author_name": "Aadil Garg",
      "author_url": "",
      "post_date": "2024-03-05T10:27:52.457000",
      "content": "<p>But how will it work with linear regression</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1909842,
      "author_name": "tdiceman",
      "author_url": "",
      "post_date": "2022-08-23T01:33:00.127000",
      "content": "<p>My guess would be that weights are kept a secret to avoid probing from strategic submissions, so that the public score provides a good indication but also doesn't give away any leak from test data. Most likely we have to rely on our own log_loss performance on a hold out set. But I could be wrong :)</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1904670,
      "author_name": "Abhishek Vijayan",
      "author_url": "",
      "post_date": "2022-08-18T11:42:42.273000",
      "content": "<p>I understood 'wi' to be weight for each class.<br>\nBut it doesn't really specify what weights are given to each of the 2 classes - CE and LAA here<br>\nMaybe they are both given 0.5 ?<br>\nClarification on this would be good</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2682440": "But how will it work with linear regression",
    "1903016": "Submissions are evaluated using a weighted multi-class logarithmic loss. This function is given under Overview->Evaluation.\nAll the parameters have been defined except 'w'. I thought it to be weights but it doesn't satisfy the equation mathematically.. Any idea what 'w' might be ?",
    "1909842": "My guess would be that weights are kept a secret to avoid probing from strategic submissions, so that the public score provides a good indication but also doesn't give away any leak from test data. Most likely we have to rely on our own log_loss performance on a hold out set. But I could be wrong :)",
    "1904670": "I understood 'wi' to be weight for each class.\nBut it doesn't really specify what weights are given to each of the 2 classes - CE and LAA here\nMaybe they are both given 0.5 ?\nClarification on this would be good"
  }
}