{
  "id": 6822,
  "title": "Why the score is tuned?",
  "url": "/competitions/yandex-personalized-web-search-challenge/discussion/6822",
  "author_name": "",
  "post_date": "2014-01-13T06:37:16.167Z",
  "votes": null,
  "comment_count": 4,
  "views": 1895,
  "content": "<p>I found all the scores were tuned after this competition over.</p>\n<p>The original baseline was 0.79056, then, it became 0.79133</p>\n<p>What happened?</p>",
  "messages": [
    {
      "id": "37331",
      "postDate": "01/13/2014 06:37:16",
      "content": "<p>I found all the scores were tuned after this competition over.</p>\n<p>The original baseline was 0.79056, then, it became 0.79133</p>\n<p>What happened?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "37332",
      "postDate": "01/13/2014 07:59:10",
      "content": "<p>Hi,</p>\n<p>the scores were not tuned: 0.79056 was the baseline score on the public leaderboard (half of the full test set), 0.79133 is the score on the private leaderboard (the full test set).</p>\n<p>J.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "37334",
      "postDate": "01/13/2014 08:52:11",
      "content": "<p>Wow! so you mean before the competition was over, online judgement only evaluate 50% of a submission data set . Am I right?</p>\n<p>That's surprising to me. Thx a lot!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "37338",
      "postDate": "01/13/2014 10:03:46",
      "content": "<p>Yes, it is like this in all competitions. The purpose is to avoid teams to overfit the public leaderboard. It was not really a problem in this competition due to the size of the dataset but, have a look at this great post in a past competition:</p>\n<p><a href=\"http://blog.kaggle.com/2012/07/06/the-dangers-of-overfitting-psychopathy-post-mortem/\">http://blog.kaggle.com/2012/07/06/the-dangers-of-overfitting-psychopathy-post-mortem/</a></p>\n<p>&nbsp; J.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "37344",
      "postDate": "01/13/2014 12:02:09",
      "content": "<p>Thank for the information!</p>\n<p>It's really nice to get the answer for you intelligent people :)&nbsp;</p>\n<p>&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 37332,
      "author_name": "jmeynet",
      "author_url": "",
      "post_date": "01/13/2014 07:59:10",
      "content": "<p>Hi,</p>\n<p>the scores were not tuned: 0.79056 was the baseline score on the public leaderboard (half of the full test set), 0.79133 is the score on the private leaderboard (the full test set).</p>\n<p>J.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 37334,
      "author_name": "ccha97u",
      "author_url": "",
      "post_date": "01/13/2014 08:52:11",
      "content": "<p>Wow! so you mean before the competition was over, online judgement only evaluate 50% of a submission data set . Am I right?</p>\n<p>That's surprising to me. Thx a lot!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 37338,
      "author_name": "jmeynet",
      "author_url": "",
      "post_date": "01/13/2014 10:03:46",
      "content": "<p>Yes, it is like this in all competitions. The purpose is to avoid teams to overfit the public leaderboard. It was not really a problem in this competition due to the size of the dataset but, have a look at this great post in a past competition:</p>\n<p><a href=\"http://blog.kaggle.com/2012/07/06/the-dangers-of-overfitting-psychopathy-post-mortem/\">http://blog.kaggle.com/2012/07/06/the-dangers-of-overfitting-psychopathy-post-mortem/</a></p>\n<p>&nbsp; J.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 37344,
      "author_name": "ccha97u",
      "author_url": "",
      "post_date": "01/13/2014 12:02:09",
      "content": "<p>Thank for the information!</p>\n<p>It's really nice to get the answer for you intelligent people :)&nbsp;</p>\n<p>&nbsp;</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "37331": "",
    "37332": "",
    "37334": "",
    "37338": "",
    "37344": ""
  },
  "source": "meta"
}