{
  "id": 10182,
  "title": "Classification label problem",
  "url": "/competitions/seizure-prediction/discussion/10182",
  "author_name": "",
  "post_date": "2014-09-01T16:59:21.287Z",
  "votes": null,
  "comment_count": 4,
  "views": 1448,
  "content": "<p>I am more from the machine learning side, so I'm not really familiar with EEG data. Any help is welcome! Anyway, am I right if I say that interictal segments(6 in total) of an hour are followed by a segment(that is not available) which should have a 0 as label? And that preictal segments(6 in total) of an hour are followed by a segment(that is not available) which should have a 1 as label.</p>\n<p>I have my doubts about it..</p>\n<p>If I am right about this, is it correct to I say that each segment of the test data should be classified as interictal(label as 0) or preictal (label as 1)?</p>\n<p>Thanks!</p>",
  "messages": [
    {
      "id": "52867",
      "postDate": "09/01/2014 16:59:21",
      "content": "<p>I am more from the machine learning side, so I'm not really familiar with EEG data. Any help is welcome! Anyway, am I right if I say that interictal segments(6 in total) of an hour are followed by a segment(that is not available) which should have a 0 as label? And that preictal segments(6 in total) of an hour are followed by a segment(that is not available) which should have a 1 as label.</p>\n<p>I have my doubts about it..</p>\n<p>If I am right about this, is it correct to I say that each segment of the test data should be classified as interictal(label as 0) or preictal (label as 1)?</p>\n<p>Thanks!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "52868",
      "postDate": "09/01/2014 17:07:30",
      "content": "<p>The evaluation metric is AUC, not classification accuracy. So, yes, it is a binary classification problem: classify as interictal 0 or preictal 1, but you probably want to rank these:</p>\n<p>test_1,0.95 #pretty sure its preictal</p>\n<p>test_2,0.5100 #not sure</p>\n<p>test_3, 0.00052 #pretty sure its interictal</p>\n<p>Or, optionally, Kaggle parser also accepts:</p>\n<p>test_1,3</p>\n<p>test_2, 2</p>\n<p>test_3, 1</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "52869",
      "postDate": "09/01/2014 17:09:57",
      "content": "<p>Thanks! got it</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "53293",
      "postDate": "09/07/2014 16:59:32",
      "content": "<p>Wait a moment. Now I'm confused.</p>\n<p>It's only the input labels that are binary, correct? The output is a probability ( look at the &quot;Submission&quot; topic).</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "53296",
      "postDate": "09/07/2014 17:53:14",
      "content": "<p>Correct! I guess</p>\n<p>So a '0' is a interictal class with probability of 1.</p>\n<p>A '1' is a preictal class with a probability of 1</p>\n<p>So if the output of your classifier gives you a '1'(preictal) with probability of 0.8, then your final answer is 0.8.</p>\n<p>If the output of your classifier gives you a '0'(interictal' with a probability of 0.8, then your final answer is a 0.2</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 52868,
      "author_name": "triskelion",
      "author_url": "",
      "post_date": "09/01/2014 17:07:30",
      "content": "<p>The evaluation metric is AUC, not classification accuracy. So, yes, it is a binary classification problem: classify as interictal 0 or preictal 1, but you probably want to rank these:</p>\n<p>test_1,0.95 #pretty sure its preictal</p>\n<p>test_2,0.5100 #not sure</p>\n<p>test_3, 0.00052 #pretty sure its interictal</p>\n<p>Or, optionally, Kaggle parser also accepts:</p>\n<p>test_1,3</p>\n<p>test_2, 2</p>\n<p>test_3, 1</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 52869,
      "author_name": "pbsciencelab",
      "author_url": "",
      "post_date": "09/01/2014 17:09:57",
      "content": "<p>Thanks! got it</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 53293,
      "author_name": "kdoniger",
      "author_url": "",
      "post_date": "09/07/2014 16:59:32",
      "content": "<p>Wait a moment. Now I'm confused.</p>\n<p>It's only the input labels that are binary, correct? The output is a probability ( look at the &quot;Submission&quot; topic).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 53296,
      "author_name": "pbsciencelab",
      "author_url": "",
      "post_date": "09/07/2014 17:53:14",
      "content": "<p>Correct! I guess</p>\n<p>So a '0' is a interictal class with probability of 1.</p>\n<p>A '1' is a preictal class with a probability of 1</p>\n<p>So if the output of your classifier gives you a '1'(preictal) with probability of 0.8, then your final answer is 0.8.</p>\n<p>If the output of your classifier gives you a '0'(interictal' with a probability of 0.8, then your final answer is a 0.2</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "52867": "",
    "52868": "",
    "52869": "",
    "53293": "",
    "53296": ""
  },
  "source": "meta"
}