{
  "id": 79980,
  "title": "A new approach - can this work?",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/79980",
  "author_name": "RVPrakash",
  "post_date": "2019-02-09T07:18:36.955000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Can we look at the target as the sum of 2 components - historic transactions + new transactions?\nThere are 13608 card_ids without new transactions and perhaps they need to be handled separately</p>\n\n<p>model 1 will predict a target score based only on historic transactions - to be trained with data without any new transactions\nmodel 2 will predict the difference between the predictions of model 1 and the actual target - to be trained with data having new transactions</p>\n\n<p>Any thoughts?  has this been tried?</p>",
  "messages": [
    {
      "id": 468577,
      "postDate": "2019-02-09T07:18:36.957Z",
      "content": "<p>Can we look at the target as the sum of 2 components - historic transactions + new transactions?\nThere are 13608 card_ids without new transactions and perhaps they need to be handled separately</p>\n\n<p>model 1 will predict a target score based only on historic transactions - to be trained with data without any new transactions\nmodel 2 will predict the difference between the predictions of model 1 and the actual target - to be trained with data having new transactions</p>\n\n<p>Any thoughts?  has this been tried?</p>",
      "rawMarkdown": "Can we look at the target as the sum of 2 components - historic transactions + new transactions?\nThere are 13608 card_ids without new transactions and perhaps they need to be handled separately\n\n\nmodel 1 will predict a target score based only on historic transactions - to be trained with data without any new transactions\nmodel 2 will predict the difference between the predictions of model 1 and the actual target - to be trained with data having new transactions\n\nAny thoughts?  has this been tried?\n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "468577": "Can we look at the target as the sum of 2 components - historic transactions + new transactions?\nThere are 13608 card_ids without new transactions and perhaps they need to be handled separately\n\n\nmodel 1 will predict a target score based only on historic transactions - to be trained with data without any new transactions\nmodel 2 will predict the difference between the predictions of model 1 and the actual target - to be trained with data having new transactions\n\nAny thoughts?  has this been tried?\n"
  }
}