{
  "id": 11513,
  "title": "Bounding Probabilities for logloss calculation",
  "url": "/competitions/avazu-ctr-prediction/discussion/11513",
  "author_name": "Nishant Kumar",
  "post_date": "2015-01-13T18:54:24.583000",
  "votes": 0,
  "comment_count": 2,
  "views": 1228,
  "content": "<p>Hi,</p>\n<p>This is a question especially for admins. The evaluation page just shows the formula for logloss but does not reveal the upper and lower bound of probabilities used for calculating logloss on leaderboard(if there is no upper and lower bound, log loss will be infinity like log(0) , so there must be some bound). Can we please know the exact equation which we are trying to optimize in this competition or point me to the link in case you have already published it. Thanks for your help !&nbsp;</p>",
  "messages": [
    {
      "id": 62714,
      "postDate": "2015-01-14T06:58:03.677Z",
      "content": "<p>The point is to prevent us from making extreme predictions - an infinite penalty if you're 100% certain that the user will do a given action (1 or 0), when it actually isn't like that and can't be like that - nature always has outliers.<br><br></p>"
    },
    {
      "id": 62700,
      "postDate": "2015-01-13T18:54:24.583Z",
      "content": "<p>Hi,</p>\n<p>This is a question especially for admins. The evaluation page just shows the formula for logloss but does not reveal the upper and lower bound of probabilities used for calculating logloss on leaderboard(if there is no upper and lower bound, log loss will be infinity like log(0) , so there must be some bound). Can we please know the exact equation which we are trying to optimize in this competition or point me to the link in case you have already published it. Thanks for your help !&nbsp;</p>"
    },
    {
      "id": 62703,
      "postDate": "2015-01-13T20:22:51.873Z",
      "rawMarkdown": "",
      "votes": 4,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 62714,
      "author_name": "danuker",
      "author_url": "",
      "post_date": "2015-01-14T06:58:03.677000",
      "content": "<p>The point is to prevent us from making extreme predictions - an infinite penalty if you're 100% certain that the user will do a given action (1 or 0), when it actually isn't like that and can't be like that - nature always has outliers.<br><br></p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 62703,
      "author_name": "",
      "author_url": "",
      "post_date": "2015-01-13T20:22:51.873000",
      "content": "",
      "votes": 4,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "62714": "",
    "62700": "",
    "62703": ""
  }
}