{
  "id": 9461,
  "title": "How were training and test customers selected ?",
  "url": "/competitions/acquire-valued-shoppers-challenge/discussion/9461",
  "author_name": "B Yang",
  "post_date": "2014-06-13T06:29:31.507000",
  "votes": 0,
  "comment_count": 0,
  "views": 742,
  "content": "<p>Hi,</p>\n\n\n<p>I did some tallying of the transaction data belonging to training and test customers. Can anyone confirm the following numbers ?</p>\n\n\n<p>For customers in trainHistory.csv:</p>\n<ul>\n<li><span style=\"line-height: 1.4\">total transactions: 187105085</span></li>\n<li><span style=\"line-height: 1.4\">transaction date range: 2012-03-02 to 2013-04-29</span></li>\n<li><span style=\"line-height: 1.4\">avg transactions per customer per day: 2.76</span></li>\n</ul>\n<p>For customers in testHistory.csv:</p>\n<ul>\n<li><span style=\"line-height: 1.4\">total transactions: 162550704</span></li>\n<li><span style=\"line-height: 1.4\">transaction date range: 2012-03-02 to 2013-07-28</span></li>\n<li><span style=\"line-height: 1.4\">avg transactions per customer per day: 2.09</span></li>\n</ul>\n<p>Here's a plot of daily transactions belonging to training and test customers:</p>\n<p><img src=\"http://i.imgur.com/3qGtekr.png\" alt=\"daily transactions\" width=\"907\" height=\"609\"></p>\n<p><br>I may have a bug in the code, but the plot looks OK, since you can see the weekly pattern and disturbances around Easter, Halloween, Christmas, and Groundhog day (I didn't know it's such a big deal). But if these numbers are correct, they clearly show training set customers do more transactions than test set customers, and they were not randomly selected subsets.</p>\n<p>So my question to competition admin is, how were training and test set customers selected ? And what caused the drop offs in transaction counts in Mar 2013 and June 2013 ?</p>",
  "messages": [
    {
      "id": 49087,
      "postDate": "2014-06-13T06:29:31.507Z",
      "content": "<p>Hi,</p>\n\n\n<p>I did some tallying of the transaction data belonging to training and test customers. Can anyone confirm the following numbers ?</p>\n\n\n<p>For customers in trainHistory.csv:</p>\n<ul>\n<li><span style=\"line-height: 1.4\">total transactions: 187105085</span></li>\n<li><span style=\"line-height: 1.4\">transaction date range: 2012-03-02 to 2013-04-29</span></li>\n<li><span style=\"line-height: 1.4\">avg transactions per customer per day: 2.76</span></li>\n</ul>\n<p>For customers in testHistory.csv:</p>\n<ul>\n<li><span style=\"line-height: 1.4\">total transactions: 162550704</span></li>\n<li><span style=\"line-height: 1.4\">transaction date range: 2012-03-02 to 2013-07-28</span></li>\n<li><span style=\"line-height: 1.4\">avg transactions per customer per day: 2.09</span></li>\n</ul>\n<p>Here's a plot of daily transactions belonging to training and test customers:</p>\n<p><img src=\"http://i.imgur.com/3qGtekr.png\" alt=\"daily transactions\" width=\"907\" height=\"609\"></p>\n<p><br>I may have a bug in the code, but the plot looks OK, since you can see the weekly pattern and disturbances around Easter, Halloween, Christmas, and Groundhog day (I didn't know it's such a big deal). But if these numbers are correct, they clearly show training set customers do more transactions than test set customers, and they were not randomly selected subsets.</p>\n<p>So my question to competition admin is, how were training and test set customers selected ? And what caused the drop offs in transaction counts in Mar 2013 and June 2013 ?</p>"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "49087": ""
  }
}