{
  "id": 10261,
  "title": "AUC sample submission ",
  "url": "/competitions/seizure-prediction/discussion/10261",
  "author_name": "",
  "post_date": "2014-09-08T09:55:57.577Z",
  "votes": null,
  "comment_count": 2,
  "views": 1283,
  "content": "<p>Still don't understand the AUC calculation or the score completely. My question, the score of .5 for the sample submission, does that mean that the test set is balanced? In other words, are the number of samples in class 0 equal to the number of samples in class 0?</p>\n<p>Or is my interpretation wrong because the score represents the AUC?</p>\n\n<p>Found that random guessing most of the time ends up with a score of .5</p>\n<p>So is the test set balanced?</p>",
  "messages": [
    {
      "id": "53334",
      "postDate": "09/08/2014 09:55:57",
      "content": "<p>Still don't understand the AUC calculation or the score completely. My question, the score of .5 for the sample submission, does that mean that the test set is balanced? In other words, are the number of samples in class 0 equal to the number of samples in class 0?</p>\n<p>Or is my interpretation wrong because the score represents the AUC?</p>\n\n<p>Found that random guessing most of the time ends up with a score of .5</p>\n<p>So is the test set balanced?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "53358",
      "postDate": "09/08/2014 18:38:26",
      "content": "<p>not necessarily, its more like the probability that you are right.</p>\n<p>for example if</p>\n<p>Ytrue = [0,0,0,1]</p>\n<p>Yestimated = [1,1,0,1]</p>\n<p>you were right half the time, but the classes aren't balanced. Its probably more complicated though.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "55133",
      "postDate": "09/24/2014 13:09:24",
      "content": "<p>1. It kinda is more complicated :-) http://en.wikipedia.org/wiki/Receiver_operating_characteristic</p>\n<p>2. AUC value tells you nothing about balance of the target values - and in fact it can be somewhat misleading, so if you're suspecting imbalance, metrics like F1 can be more&nbsp;informative.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 53358,
      "author_name": "franklyn",
      "author_url": "",
      "post_date": "09/08/2014 18:38:26",
      "content": "<p>not necessarily, its more like the probability that you are right.</p>\n<p>for example if</p>\n<p>Ytrue = [0,0,0,1]</p>\n<p>Yestimated = [1,1,0,1]</p>\n<p>you were right half the time, but the classes aren't balanced. Its probably more complicated though.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 55133,
      "author_name": "konradb",
      "author_url": "",
      "post_date": "09/24/2014 13:09:24",
      "content": "<p>1. It kinda is more complicated :-) http://en.wikipedia.org/wiki/Receiver_operating_characteristic</p>\n<p>2. AUC value tells you nothing about balance of the target values - and in fact it can be somewhat misleading, so if you're suspecting imbalance, metrics like F1 can be more&nbsp;informative.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "53334": "",
    "53358": "",
    "55133": ""
  },
  "source": "meta"
}