{
  "id": 710,
  "title": "some makes/models removed from training",
  "url": "/competitions/ClaimPredictionChallenge/discussion/710",
  "author_name": "Allstate Insurance Company",
  "post_date": "2011-07-14T12:39:28.910000",
  "votes": 0,
  "comment_count": 0,
  "views": 1669,
  "content": "<p>Hi! There's an important feature of the dataset that I forgot to include in the data description:</p>\r\n<p>One interesting/difficult problem for us is coming up with predictions each year for 'new' makes and models/submodels that didn't exist in previous years. Since the prediction algorithm we decide to use may be in use for a period of years, the problem of\r\n predicting new makes and models is actually more important than it would appear from a testing dataset that is just one or two years past the training dataset.</p>\r\n<p>For this reason, some models/submodels have been removed from the training dataset. I'm sorry if that has caused anyone confusion.</p>",
  "messages": [
    {
      "id": 4599,
      "postDate": "2011-07-14T12:39:28.910Z",
      "content": "<p>Hi! There's an important feature of the dataset that I forgot to include in the data description:</p>\r\n<p>One interesting/difficult problem for us is coming up with predictions each year for 'new' makes and models/submodels that didn't exist in previous years. Since the prediction algorithm we decide to use may be in use for a period of years, the problem of\r\n predicting new makes and models is actually more important than it would appear from a testing dataset that is just one or two years past the training dataset.</p>\r\n<p>For this reason, some models/submodels have been removed from the training dataset. I'm sorry if that has caused anyone confusion.</p>"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "4599": ""
  }
}