{
  "id": 9813,
  "title": "Avito Best Prediction",
  "url": "/competitions/avito-prohibited-content/discussion/9813",
  "author_name": "",
  "post_date": "2014-07-19T00:59:55.440Z",
  "votes": null,
  "comment_count": 2,
  "views": 1327,
  "content": "<p>Hi,</p>\n<p>Any idea what are the parameters settings, and the number of samples etc for avito best benchmark? I can&nbsp;not really change all the parameters to see how the best prediction benchmark is being achieved, because the number of entries per day is limited.</p>\n\n\n<p>Sayan</p>",
  "messages": [
    {
      "id": "50886",
      "postDate": "07/19/2014 00:59:55",
      "content": "<p>Hi,</p>\n<p>Any idea what are the parameters settings, and the number of samples etc for avito best benchmark? I can&nbsp;not really change all the parameters to see how the best prediction benchmark is being achieved, because the number of entries per day is limited.</p>\n\n\n<p>Sayan</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "50890",
      "postDate": "07/19/2014 07:25:35",
      "content": "<p>How about tuning your model parameters with K-fold cross validation?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "50891",
      "postDate": "07/19/2014 08:27:08",
      "content": "<p>I can certainly do that, but the problem is the accuracy in k-fold validation will not vary as much as the MAP, since the accuracy/TPR-FPR&nbsp;and other measures are just average on positive or negative outcimes, but MAP depends on the positions of the positive outcomes (assuming the illicit ads has positive labels). So the accurate tuning of parameters are more difficult using k-fold validation.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 50890,
      "author_name": "apophenia",
      "author_url": "",
      "post_date": "07/19/2014 07:25:35",
      "content": "<p>How about tuning your model parameters with K-fold cross validation?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 50891,
      "author_name": "sayan21",
      "author_url": "",
      "post_date": "07/19/2014 08:27:08",
      "content": "<p>I can certainly do that, but the problem is the accuracy in k-fold validation will not vary as much as the MAP, since the accuracy/TPR-FPR&nbsp;and other measures are just average on positive or negative outcimes, but MAP depends on the positions of the positive outcomes (assuming the illicit ads has positive labels). So the accurate tuning of parameters are more difficult using k-fold validation.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "50886": "",
    "50890": "",
    "50891": ""
  },
  "source": "meta"
}