{
  "id": 10185,
  "title": "number of test clips",
  "url": "/competitions/seizure-prediction/discussion/10185",
  "author_name": "",
  "post_date": "2014-09-01T23:48:28.847Z",
  "votes": null,
  "comment_count": 1,
  "views": 838,
  "content": "<p>Each subject has a different number of test examples, which can bias the evaluation metrics towards the subjects with most examples. Does the metric account for the difference (for example, by taking an average of AUCs calculated for each subject separately)?</p>\n\n<p>Bartosz</p>",
  "messages": [
    {
      "id": "52889",
      "postDate": "09/01/2014 23:48:28",
      "content": "<p>Each subject has a different number of test examples, which can bias the evaluation metrics towards the subjects with most examples. Does the metric account for the difference (for example, by taking an average of AUCs calculated for each subject separately)?</p>\n\n<p>Bartosz</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "52912",
      "postDate": "09/02/2014 10:32:41",
      "content": "<p>Not only towards the subjects with most examples, but also with respect to the different ratios of preictal/interictal test examples in each subject.</p>\n<p>I understand that the AUC is computed globally, not by averaging 7 per-subject AUCs. But I agree with you: intuitively, the average would be a fairer measure of performance.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 52912,
      "author_name": "joseleiva",
      "author_url": "",
      "post_date": "09/02/2014 10:32:41",
      "content": "<p>Not only towards the subjects with most examples, but also with respect to the different ratios of preictal/interictal test examples in each subject.</p>\n<p>I understand that the AUC is computed globally, not by averaging 7 per-subject AUCs. But I agree with you: intuitively, the average would be a fairer measure of performance.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "52889": "",
    "52912": ""
  },
  "source": "meta"
}