{
  "id": 6161,
  "title": "Normalized Discounted Cumulative Gain  - Three versions ?",
  "url": "/competitions/yandex-personalized-web-search-challenge/discussion/6161",
  "author_name": "",
  "post_date": "2013-10-30T17:40:50.540Z",
  "votes": null,
  "comment_count": 2,
  "views": 1351,
  "content": "<p>In reading over the Wikipedia entry for NDCG, it turns out there are two formulations for Discounted Cumulative Gain.&nbsp;</p>\n<p>http://en.wikipedia.org/wiki/Discounted_cumulative_gain#Discounted_Cumulative_Gain</p>\n<p>There is also a third version which used natural logarithm on the second formula.</p>\n<p>(1)&nbsp; dcg(p) = rel(1) + sum(rel(i)/log2(i))&nbsp; for i in 2:p<br>(2)&nbsp; dcg(p) = sum(((2**rel(i))-1)/(log2(i+1)))&nbsp;&nbsp; for i in 1:p<br>(3)&nbsp; dcg(p) = sum(((2**rel(i))-1)/(loge(i+1)))&nbsp;&nbsp; for i in 1:p</p>\n<p>&nbsp;</p>\n<p>Which of the three versions for DCG will be used by Yandex for scoring?&nbsp;</p>\n<p>Thanks!</p>\n<p>&nbsp;</p>",
  "messages": [
    {
      "id": "32892",
      "postDate": "10/30/2013 17:40:50",
      "content": "<p>In reading over the Wikipedia entry for NDCG, it turns out there are two formulations for Discounted Cumulative Gain.&nbsp;</p>\n<p>http://en.wikipedia.org/wiki/Discounted_cumulative_gain#Discounted_Cumulative_Gain</p>\n<p>There is also a third version which used natural logarithm on the second formula.</p>\n<p>(1)&nbsp; dcg(p) = rel(1) + sum(rel(i)/log2(i))&nbsp; for i in 2:p<br>(2)&nbsp; dcg(p) = sum(((2**rel(i))-1)/(log2(i+1)))&nbsp;&nbsp; for i in 1:p<br>(3)&nbsp; dcg(p) = sum(((2**rel(i))-1)/(loge(i+1)))&nbsp;&nbsp; for i in 1:p</p>\n<p>&nbsp;</p>\n<p>Which of the three versions for DCG will be used by Yandex for scoring?&nbsp;</p>\n<p>Thanks!</p>\n<p>&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "32893",
      "postDate": "10/30/2013 18:02:53",
      "content": "<p>Hello,</p>\n<p>We use the default implementation provided by Kaggle. The description and an example implementation can be found on Kaggle's <a href=\"https://www.kaggle.com/wiki/NormalizedDiscountedCumulativeGain\">wiki</a>.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "32894",
      "postDate": "10/30/2013 18:08:08",
      "content": "<p>Thanks !</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 32893,
      "author_name": "eugene1751",
      "author_url": "",
      "post_date": "10/30/2013 18:02:53",
      "content": "<p>Hello,</p>\n<p>We use the default implementation provided by Kaggle. The description and an example implementation can be found on Kaggle's <a href=\"https://www.kaggle.com/wiki/NormalizedDiscountedCumulativeGain\">wiki</a>.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 32894,
      "author_name": "geringer",
      "author_url": "",
      "post_date": "10/30/2013 18:08:08",
      "content": "<p>Thanks !</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "32892": "",
    "32893": "",
    "32894": ""
  },
  "source": "meta"
}