{
  "id": 11539,
  "title": "Question about Evaluation, thanks!",
  "url": "/competitions/avazu-ctr-prediction/discussion/11539",
  "author_name": "",
  "post_date": "2015-01-17T19:00:21.880Z",
  "votes": null,
  "comment_count": 2,
  "views": 970,
  "content": "<p>Dear&nbsp;competition participants, competition administrators,</p>\n<p>Sorry to bother you guys. I know the question is silly. Could anyone&nbsp;please explain how to evaluate the results? What does it mean by using a double sum here?</p>\n<p>https://www.kaggle.com/wiki/LogarithmicLoss</p>\n<p>Thank you in advance. :)</p>\n\n<p>Best,</p>\n<p>Xiao</p>",
  "messages": [
    {
      "id": "62841",
      "postDate": "01/17/2015 19:00:21",
      "content": "<p>Dear&nbsp;competition participants, competition administrators,</p>\n<p>Sorry to bother you guys. I know the question is silly. Could anyone&nbsp;please explain how to evaluate the results? What does it mean by using a double sum here?</p>\n<p>https://www.kaggle.com/wiki/LogarithmicLoss</p>\n<p>Thank you in advance. :)</p>\n\n<p>Best,</p>\n<p>Xiao</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "62844",
      "postDate": "01/17/2015 21:10:21",
      "content": "<p>In the expression:</p>\n<p>$$log loss = -\\frac{1}{N}\\sum_{i=1}^N\\sum_{j=1}^My_{ij}\\log(p_{ij})$$ &nbsp;</p>\n<p>the sum over 1..N is over the examples and the sum over 1..M is over the classes. This is a form that is valid for any number of classes. In this competition, there are only two classes, so this is equivalent to:</p>\n<p>$$log loss = -\\frac{1}{N}\\sum_{i=1}^N[I\\{y_i=0\\}\\log(p(y_i=0)) + I\\{y_i=1\\}\\log(p(y_i=1))]$$ &nbsp;&nbsp;</p>\n<p>where I{.} denotes an indicator function and the p(y=.) are your predicted probabilities.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "62846",
      "postDate": "01/17/2015 22:50:03",
      "content": "<p>Thank you&nbsp;David! It is really helpful. Now I know how to calculate it.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 62844,
      "author_name": "davidthaler",
      "author_url": "",
      "post_date": "01/17/2015 21:10:21",
      "content": "<p>In the expression:</p>\n<p>$$log loss = -\\frac{1}{N}\\sum_{i=1}^N\\sum_{j=1}^My_{ij}\\log(p_{ij})$$ &nbsp;</p>\n<p>the sum over 1..N is over the examples and the sum over 1..M is over the classes. This is a form that is valid for any number of classes. In this competition, there are only two classes, so this is equivalent to:</p>\n<p>$$log loss = -\\frac{1}{N}\\sum_{i=1}^N[I\\{y_i=0\\}\\log(p(y_i=0)) + I\\{y_i=1\\}\\log(p(y_i=1))]$$ &nbsp;&nbsp;</p>\n<p>where I{.} denotes an indicator function and the p(y=.) are your predicted probabilities.&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 62846,
      "author_name": "xwangstat",
      "author_url": "",
      "post_date": "01/17/2015 22:50:03",
      "content": "<p>Thank you&nbsp;David! It is really helpful. Now I know how to calculate it.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "62841": "",
    "62844": "",
    "62846": ""
  },
  "source": "meta"
}