{
  "id": 56443,
  "title": "Adversarial Validation within Test Data Makes Sense ?",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/56443",
  "author_name": "",
  "post_date": "2018-05-09T19:45:24.587836600Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Towards the end of the competition , we started ensembling (Before DKD (Dirt's Kernel Disaster) . To check if we are overfitting , i ran adversarial validation labelling test's Hour no.4 as 1 and rest as 0 and got a AUC ~ 0.57.</p>\n\n<p>Now my guess was , this means the test and train data are not that different .</p>\n\n<p>What is the correct way to interpret it ? does it makes sense or its nonsense.</p>\n\n<p>Obviously , this can be done only when we know what does PB data belongs to </p>",
  "messages": [
    {
      "id": "326478",
      "postDate": "05/09/2018 19:45:24",
      "content": "<p>Towards the end of the competition , we started ensembling (Before DKD (Dirt's Kernel Disaster) . To check if we are overfitting , i ran adversarial validation labelling test's Hour no.4 as 1 and rest as 0 and got a AUC ~ 0.57.</p>\n\n<p>Now my guess was , this means the test and train data are not that different .</p>\n\n<p>What is the correct way to interpret it ? does it makes sense or its nonsense.</p>\n\n<p>Obviously , this can be done only when we know what does PB data belongs to </p>",
      "rawMarkdown": "Towards the end of the competition , we started ensembling (Before DKD (Dirt's Kernel Disaster) . To check if we are overfitting , i ran adversarial validation labelling test's Hour no.4 as 1 and rest as 0 and got a AUC ~ 0.57.\n\nNow my guess was , this means the test and train data are not that different .\n\nWhat is the correct way to interpret it ? does it makes sense or its nonsense.\n\nObviously , this can be done only when we know what does PB data belongs to",
      "votes": null
    },
    {
      "id": "326515",
      "postDate": "05/09/2018 21:11:15",
      "content": "<p>0.57 auc is significant.  </p>",
      "rawMarkdown": "0.57 auc is significant.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 326515,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "05/09/2018 21:11:15",
      "content": "<p>0.57 auc is significant.  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "326478": "Towards the end of the competition , we started ensembling (Before DKD (Dirt's Kernel Disaster) . To check if we are overfitting , i ran adversarial validation labelling test's Hour no.4 as 1 and rest as 0 and got a AUC ~ 0.57.\n\nNow my guess was , this means the test and train data are not that different .\n\nWhat is the correct way to interpret it ? does it makes sense or its nonsense.\n\nObviously , this can be done only when we know what does PB data belongs to",
    "326515": "0.57 auc is significant."
  },
  "source": "meta"
}