{
  "id": 10812,
  "title": "Description of named variables",
  "url": "/competitions/avazu-ctr-prediction/discussion/10812",
  "author_name": "",
  "post_date": "2014-11-01T04:00:38.977Z",
  "votes": null,
  "comment_count": 1,
  "views": 934,
  "content": "<p>Hello all,</p>\n<p>I'm wondering if we can get more specific description of the named variables, which I believe is vary important for us to tune the model.</p>\n<p>Without further description and if the named variables are understood literally, I'm very confused with the data provided. For example, in the training data, the same <strong>device_id&nbsp;</strong>has different <strong>device_make</strong>, <strong>device_model</strong> and <strong>device_os</strong>. But how to understand this? One device should have just one id, one operation system, and is belonging to one kind of model.</p>\n<p>Thanks!</p>",
  "messages": [
    {
      "id": "57150",
      "postDate": "11/01/2014 04:00:38",
      "content": "<p>Hello all,</p>\n<p>I'm wondering if we can get more specific description of the named variables, which I believe is vary important for us to tune the model.</p>\n<p>Without further description and if the named variables are understood literally, I'm very confused with the data provided. For example, in the training data, the same <strong>device_id&nbsp;</strong>has different <strong>device_make</strong>, <strong>device_model</strong> and <strong>device_os</strong>. But how to understand this? One device should have just one id, one operation system, and is belonging to one kind of model.</p>\n<p>Thanks!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "57152",
      "postDate": "11/01/2014 06:36:37",
      "content": "<p>Yes, there are 0.1% noise in the origianl data with respect to&nbsp;device_id +&nbsp;device_make, device_model, .etc. Whether to clean it before modeling or not depend on your choice.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 57152,
      "author_name": "stevewang0",
      "author_url": "",
      "post_date": "11/01/2014 06:36:37",
      "content": "<p>Yes, there are 0.1% noise in the origianl data with respect to&nbsp;device_id +&nbsp;device_make, device_model, .etc. Whether to clean it before modeling or not depend on your choice.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "57150": "",
    "57152": ""
  },
  "source": "meta"
}