{
  "id": 8087,
  "title": ">1 model",
  "url": "/competitions/acquire-valued-shoppers-challenge/discussion/8087",
  "author_name": "",
  "post_date": "2014-05-08T00:49:15.587Z",
  "votes": null,
  "comment_count": 3,
  "views": 1393,
  "content": "<p>I am thinking of building 2 different models based customers who have purchased say the specific Company+Category+Brand items as opposed to one who hasn't. In the train set, customers who have done this are approx.1.6 times more likely to be a repeater. Without doing the detailed analysis yet, I am thinking that there is bound to be different behaviours between these 2 segments that may not be sufficient for one single model to accomodate. Maybe an opportunity to boost predictive power with a multitude of models.&nbsp;</p>\n<p>Interested in finding out if others have had similar thoughts or are already doing this. If so what sort of performance results are you getting?&nbsp;</p>",
  "messages": [
    {
      "id": "44222",
      "postDate": "05/08/2014 00:49:15",
      "content": "<p>I am thinking of building 2 different models based customers who have purchased say the specific Company+Category+Brand items as opposed to one who hasn't. In the train set, customers who have done this are approx.1.6 times more likely to be a repeater. Without doing the detailed analysis yet, I am thinking that there is bound to be different behaviours between these 2 segments that may not be sufficient for one single model to accomodate. Maybe an opportunity to boost predictive power with a multitude of models.&nbsp;</p>\n<p>Interested in finding out if others have had similar thoughts or are already doing this. If so what sort of performance results are you getting?&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "44270",
      "postDate": "05/08/2014 17:25:43",
      "content": "<p>In theory, if you gave that variable to any tree-based model (has bought&nbsp;Company+Category+Brand in transactions), it should be able to model the 2 groups separately if needed.</p>\n\n<p>In practice, I'm not sure. &nbsp;It might be worth trying.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "49104",
      "postDate": "06/13/2014 14:54:11",
      "content": "<p>any thought on this idea? 2 models vs single model with an explicit variable? it would be interesting to know different&nbsp;views.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "49122",
      "postDate": "06/13/2014 18:42:32",
      "content": "<p>You could go further than that. You could also split the data in people who buy much and people who buy less (rich and poor people?). Or people who buy often and people who visit the shop only a few times each year. Only the&nbsp;results can tell you if it is&nbsp;a good idea.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 44270,
      "author_name": "zachmayer",
      "author_url": "",
      "post_date": "05/08/2014 17:25:43",
      "content": "<p>In theory, if you gave that variable to any tree-based model (has bought&nbsp;Company+Category+Brand in transactions), it should be able to model the 2 groups separately if needed.</p>\n\n<p>In practice, I'm not sure. &nbsp;It might be worth trying.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 49104,
      "author_name": "anshul",
      "author_url": "",
      "post_date": "06/13/2014 14:54:11",
      "content": "<p>any thought on this idea? 2 models vs single model with an explicit variable? it would be interesting to know different&nbsp;views.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 49122,
      "author_name": "jost54",
      "author_url": "",
      "post_date": "06/13/2014 18:42:32",
      "content": "<p>You could go further than that. You could also split the data in people who buy much and people who buy less (rich and poor people?). Or people who buy often and people who visit the shop only a few times each year. Only the&nbsp;results can tell you if it is&nbsp;a good idea.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "44222": "",
    "44270": "",
    "49104": "",
    "49122": ""
  },
  "source": "meta"
}