{
  "id": 10484,
  "title": "Questions about the test set",
  "url": "/competitions/seizure-prediction/discussion/10484",
  "author_name": "",
  "post_date": "2014-09-29T18:34:32.040Z",
  "votes": 1,
  "comment_count": 2,
  "views": 1114,
  "content": "<p>1) Was the Train/Test split random?&nbsp;</p>\n<p>2) Was&nbsp;the Public/Private split random (per file or per hour)?</p>\n<p>3) Does&nbsp;the <em>&quot;time machine rule&quot;</em> apply to this competition too?</p>\n<p>http://www.kaggle.com/c/seizure-detection/forums/t/8347/using-test-set-distribution</p>\n<p>Edit:</p>\n<p>1) The train/test split does not seem to be random:</p>",
  "messages": [
    {
      "id": "55399",
      "postDate": "09/29/2014 18:34:32",
      "content": "<p>1) Was the Train/Test split random?&nbsp;</p>\n<p>2) Was&nbsp;the Public/Private split random (per file or per hour)?</p>\n<p>3) Does&nbsp;the <em>&quot;time machine rule&quot;</em> apply to this competition too?</p>\n<p>http://www.kaggle.com/c/seizure-detection/forums/t/8347/using-test-set-distribution</p>\n<p>Edit:</p>\n<p>1) The train/test split does not seem to be random:</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "55430",
      "postDate": "09/30/2014 14:19:18",
      "content": "<p>I am seeing very different histograms of predicted preictal probability on&nbsp;the test data vs my CV for Dog_3, Dog_5, and the human patients.&nbsp;Don't know if it's real or a modeling error.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "55432",
      "postDate": "09/30/2014 16:42:33",
      "content": "<p>Emolson, I have the same problem, and it seems that we are not the only one. Cross validation results are not consistent with test results, and so it is possible to observe very different histograms. I have problems with Dog_1 besides Dog_3, Dog_5 and the human patients.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 55430,
      "author_name": "emolson",
      "author_url": "",
      "post_date": "09/30/2014 14:19:18",
      "content": "<p>I am seeing very different histograms of predicted preictal probability on&nbsp;the test data vs my CV for Dog_3, Dog_5, and the human patients.&nbsp;Don't know if it's real or a modeling error.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 55432,
      "author_name": "pakozm",
      "author_url": "",
      "post_date": "09/30/2014 16:42:33",
      "content": "<p>Emolson, I have the same problem, and it seems that we are not the only one. Cross validation results are not consistent with test results, and so it is possible to observe very different histograms. I have problems with Dog_1 besides Dog_3, Dog_5 and the human patients.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "55399": "",
    "55430": "",
    "55432": ""
  },
  "source": "meta"
}