{
  "id": 16215,
  "title": "AUC Question Regarding Transformation",
  "url": "/competitions/dato-native/discussion/16215",
  "author_name": "",
  "post_date": "2015-08-30T04:14:37.057Z",
  "votes": null,
  "comment_count": 2,
  "views": 664,
  "content": "<p>I have a question about the <a href=\"https://www.kaggle.com/wiki/AUC\">AUC</a> calculation used in scoring.  Let's say I make predictions for the sample data and the resulting average of those predictions is 0.15.  Now the average of the training data is around .09.  Is it true that based on how the AUC is calculated and used, it would be worthless to do a transformation to adjust my predictions to the average of the training data to try to better match it?  Is that one of benefits of this method (like as discussed in the <a href=\"https://www.youtube.com/watch?v=OAl6eAyP-yo&feature=youtu.be\">linked video</a> from the wiki)?</p>\n\n<p>Thanks in advance.</p>",
  "messages": [
    {
      "id": "90925",
      "postDate": "08/30/2015 04:14:37",
      "content": "<p>I have a question about the <a href=\"https://www.kaggle.com/wiki/AUC\">AUC</a> calculation used in scoring.  Let's say I make predictions for the sample data and the resulting average of those predictions is 0.15.  Now the average of the training data is around .09.  Is it true that based on how the AUC is calculated and used, it would be worthless to do a transformation to adjust my predictions to the average of the training data to try to better match it?  Is that one of benefits of this method (like as discussed in the <a href=\"https://www.youtube.com/watch?v=OAl6eAyP-yo&feature=youtu.be\">linked video</a> from the wiki)?</p>\n\n<p>Thanks in advance.</p>",
      "rawMarkdown": "I have a question about the [AUC](https://www.kaggle.com/wiki/AUC) calculation used in scoring.  Let's say I make predictions for the sample data and the resulting average of those predictions is 0.15.  Now the average of the training data is around .09.  Is it true that based on how the AUC is calculated and used, it would be worthless to do a transformation to adjust my predictions to the average of the training data to try to better match it?  Is that one of benefits of this method (like as discussed in the [linked video](https://www.youtube.com/watch?v=OAl6eAyP-yo&feature=youtu.be) from the wiki)?\r\n\r\nThanks in advance.",
      "votes": null
    },
    {
      "id": "90971",
      "postDate": "08/30/2015 22:39:27",
      "content": "<p>AUC is a rank-order-only metric. The mean prediction does not matter.</p>",
      "rawMarkdown": "AUC is a rank-order-only metric. The mean prediction does not matter.",
      "votes": null
    },
    {
      "id": "91205",
      "postDate": "09/01/2015 15:48:27",
      "content": "<p>Mean predictions doesn't matter.\nProbability magnitude doesn't matter.\nOnly the rank matters.</p>\n\n<p>I tried my best to explain AUC intuitively here. Go through this. <a href=\"https://www.kaggle.com/c/springleaf-marketing-response/forums/t/16055/evaluation-example/90114#post90114\">https://www.kaggle.com/c/springleaf-marketing-response/forums/t/16055/evaluation-example/90114#post90114</a></p>",
      "rawMarkdown": "Mean predictions doesn't matter.\r\nProbability magnitude doesn't matter.\r\nOnly the rank matters.\r\n\r\nI tried my best to explain AUC intuitively here. Go through this. https://www.kaggle.com/c/springleaf-marketing-response/forums/t/16055/evaluation-example/90114#post90114",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 90971,
      "author_name": "davidthaler",
      "author_url": "",
      "post_date": "08/30/2015 22:39:27",
      "content": "<p>AUC is a rank-order-only metric. The mean prediction does not matter.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 91205,
      "author_name": "phanisrikanth",
      "author_url": "",
      "post_date": "09/01/2015 15:48:27",
      "content": "<p>Mean predictions doesn't matter.\nProbability magnitude doesn't matter.\nOnly the rank matters.</p>\n\n<p>I tried my best to explain AUC intuitively here. Go through this. <a href=\"https://www.kaggle.com/c/springleaf-marketing-response/forums/t/16055/evaluation-example/90114#post90114\">https://www.kaggle.com/c/springleaf-marketing-response/forums/t/16055/evaluation-example/90114#post90114</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "90925": "I have a question about the [AUC](https://www.kaggle.com/wiki/AUC) calculation used in scoring.  Let's say I make predictions for the sample data and the resulting average of those predictions is 0.15.  Now the average of the training data is around .09.  Is it true that based on how the AUC is calculated and used, it would be worthless to do a transformation to adjust my predictions to the average of the training data to try to better match it?  Is that one of benefits of this method (like as discussed in the [linked video](https://www.youtube.com/watch?v=OAl6eAyP-yo&feature=youtu.be) from the wiki)?\r\n\r\nThanks in advance.",
    "90971": "AUC is a rank-order-only metric. The mean prediction does not matter.",
    "91205": "Mean predictions doesn't matter.\r\nProbability magnitude doesn't matter.\r\nOnly the rank matters.\r\n\r\nI tried my best to explain AUC intuitively here. Go through this. https://www.kaggle.com/c/springleaf-marketing-response/forums/t/16055/evaluation-example/90114#post90114"
  },
  "source": "meta"
}