{
  "id": 11178,
  "title": "Possible data leakage",
  "url": "/competitions/inria-bci-challenge/discussion/11178",
  "author_name": "",
  "post_date": "2014-12-08T22:02:36.373Z",
  "votes": 4,
  "comment_count": 3,
  "views": 1697,
  "content": "<p>It is possible to determine the labels that the Data Providers error detection classifier generated for session 5's by using the time between feedback events. Their classifier was trained on sessions 1-4 for the test subjects so it contains information we don't have access to otherwise. Using this gives me an extra .05 on four fold CV.</p>",
  "messages": [
    {
      "id": "59838",
      "postDate": "12/08/2014 22:02:36",
      "content": "<p>It is possible to determine the labels that the Data Providers error detection classifier generated for session 5's by using the time between feedback events. Their classifier was trained on sessions 1-4 for the test subjects so it contains information we don't have access to otherwise. Using this gives me an extra .05 on four fold CV.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "63778",
      "postDate": "02/08/2015 19:15:17",
      "content": "<p>I am doing extremely good on train set but on submission I am doing the worst .&nbsp;</p>\n<p>Can you tell me more about the data leakage ? particularly in the problem at hand</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "63887",
      "postDate": "02/10/2015 07:34:14",
      "content": "<p>In the experiment design, there is a bit more&nbsp;time&nbsp;after the last letter in&nbsp;a word.&nbsp;Perhaps that is what you are picking up with the time between feedback events. If so, I mightn't&nbsp;call it a data leakage.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "63889",
      "postDate": "02/10/2015 07:59:23",
      "content": "<p>I accounted for the extra time at the end of the words. Additionally I accounted for random variation in the time between flashing and presentation of selected letter. The extra time only shows up in session five when error correction is performed. Also there is the increase in AUC from using the information.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 63778,
      "author_name": "ydhiman",
      "author_url": "",
      "post_date": "02/08/2015 19:15:17",
      "content": "<p>I am doing extremely good on train set but on submission I am doing the worst .&nbsp;</p>\n<p>Can you tell me more about the data leakage ? particularly in the problem at hand</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 63887,
      "author_name": "danielyoo",
      "author_url": "",
      "post_date": "02/10/2015 07:34:14",
      "content": "<p>In the experiment design, there is a bit more&nbsp;time&nbsp;after the last letter in&nbsp;a word.&nbsp;Perhaps that is what you are picking up with the time between feedback events. If so, I mightn't&nbsp;call it a data leakage.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 63889,
      "author_name": "devinanzelmo",
      "author_url": "",
      "post_date": "02/10/2015 07:59:23",
      "content": "<p>I accounted for the extra time at the end of the words. Additionally I accounted for random variation in the time between flashing and presentation of selected letter. The extra time only shows up in session five when error correction is performed. Also there is the increase in AUC from using the information.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "59838": "",
    "63778": "",
    "63887": "",
    "63889": ""
  },
  "source": "meta"
}