{
  "id": 42587,
  "title": "Sample submission users with no data",
  "url": "/competitions/kkbox-churn-prediction-challenge/discussion/42587",
  "author_name": "",
  "post_date": "2017-11-01T20:50:44.721850100Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have noticed that some of the test members do not have any data on the user_logs, transactions or members datasets.  Examples of this users are: ls/2eT9UWbyV4tNdfZ4MFgUVY0c4Rx5WM1uLWVdjfVs=\nI8ovHkG/uALsd3JaclAD/aCCmxSBlkbo7YwTCJlpRGU=</p>\n\n<p>How do you deal with this users? Am i missing something?</p>",
  "messages": [
    {
      "id": "238659",
      "postDate": "11/01/2017 20:50:44",
      "content": "<p>I have noticed that some of the test members do not have any data on the user_logs, transactions or members datasets.  Examples of this users are: ls/2eT9UWbyV4tNdfZ4MFgUVY0c4Rx5WM1uLWVdjfVs=\nI8ovHkG/uALsd3JaclAD/aCCmxSBlkbo7YwTCJlpRGU=</p>\n\n<p>How do you deal with this users? Am i missing something?</p>",
      "rawMarkdown": "I have noticed that some of the test members do not have any data on the user_logs, transactions or members datasets.  Examples of this users are: ls/2eT9UWbyV4tNdfZ4MFgUVY0c4Rx5WM1uLWVdjfVs=\nI8ovHkG/uALsd3JaclAD/aCCmxSBlkbo7YwTCJlpRGU=\n\nHow do you deal with this users? Am i missing something?",
      "votes": null
    },
    {
      "id": "239015",
      "postDate": "11/02/2017 14:51:01",
      "content": "<p>We have no transactions/log history after 2017-02-28. <br>\nMost likely these users began WSDM service after this date -- unless we have missing (dirty) data. <br>\n2527 users are in test but have no transactions <br>\n1980 of these users are in the members table (547 are not) <br>\n' ' ' ' <br>\nNow using the leaked data we can deduce: <br>\n998 of the 'no transaction users' are part of the leaked data <br>\nOf these, 546 (54%) of them churned. <br>\n' ' ' ' <br>\nI have not dealt with this yet. My first approach would be to identify similar users in the training set and either build a separate model for them, or include an attribute that indicates they are a 'no_transaction' user. <br>\nIf there is no data at all for the user, I suppose it is best to predict them separately, according to basic statistical analysis like the one I did above; however, I will have to put in some thought into what the submitted probability will be! <br>\nI hope this helps!</p>",
      "rawMarkdown": "We have no transactions/log history after 2017-02-28.  \nMost likely these users began WSDM service after this date -- unless we have missing (dirty) data.  \n2527 users are in test but have no transactions  \n1980 of these users are in the members table (547 are not)  \n' ' ' '  \nNow using the leaked data we can deduce:  \n998 of the 'no transaction users' are part of the leaked data  \nOf these, 546 (54%) of them churned.  \n' ' ' '   \nI have not dealt with this yet. My first approach would be to identify similar users in the training set and either build a separate model for them, or include an attribute that indicates they are a 'no_transaction' user.  \nIf there is no data at all for the user, I suppose it is best to predict them separately, according to basic statistical analysis like the one I did above; however, I will have to put in some thought into what the submitted probability will be!  \nI hope this helps!",
      "votes": null
    },
    {
      "id": "239156",
      "postDate": "11/02/2017 19:34:27",
      "content": "<p>Thank you for your answer, it's good to know I wasn't wrong about those users and there is indeed no data attached to them. Hopefully when admins release new test data, we can make some inferences with these users :)</p>",
      "rawMarkdown": "Thank you for your answer, it's good to know I wasn't wrong about those users and there is indeed no data attached to them. Hopefully when admins release new test data, we can make some inferences with these users :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 239015,
      "author_name": "npa02012",
      "author_url": "",
      "post_date": "11/02/2017 14:51:01",
      "content": "<p>We have no transactions/log history after 2017-02-28. <br>\nMost likely these users began WSDM service after this date -- unless we have missing (dirty) data. <br>\n2527 users are in test but have no transactions <br>\n1980 of these users are in the members table (547 are not) <br>\n' ' ' ' <br>\nNow using the leaked data we can deduce: <br>\n998 of the 'no transaction users' are part of the leaked data <br>\nOf these, 546 (54%) of them churned. <br>\n' ' ' ' <br>\nI have not dealt with this yet. My first approach would be to identify similar users in the training set and either build a separate model for them, or include an attribute that indicates they are a 'no_transaction' user. <br>\nIf there is no data at all for the user, I suppose it is best to predict them separately, according to basic statistical analysis like the one I did above; however, I will have to put in some thought into what the submitted probability will be! <br>\nI hope this helps!</p>",
      "votes": null,
      "replies": [
        {
          "id": 239156,
          "author_name": "galeos9",
          "author_url": "",
          "post_date": "11/02/2017 19:34:27",
          "content": "<p>Thank you for your answer, it's good to know I wasn't wrong about those users and there is indeed no data attached to them. Hopefully when admins release new test data, we can make some inferences with these users :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "238659": "I have noticed that some of the test members do not have any data on the user_logs, transactions or members datasets.  Examples of this users are: ls/2eT9UWbyV4tNdfZ4MFgUVY0c4Rx5WM1uLWVdjfVs=\nI8ovHkG/uALsd3JaclAD/aCCmxSBlkbo7YwTCJlpRGU=\n\nHow do you deal with this users? Am i missing something?",
    "239015": "We have no transactions/log history after 2017-02-28.  \nMost likely these users began WSDM service after this date -- unless we have missing (dirty) data.  \n2527 users are in test but have no transactions  \n1980 of these users are in the members table (547 are not)  \n' ' ' '  \nNow using the leaked data we can deduce:  \n998 of the 'no transaction users' are part of the leaked data  \nOf these, 546 (54%) of them churned.  \n' ' ' '   \nI have not dealt with this yet. My first approach would be to identify similar users in the training set and either build a separate model for them, or include an attribute that indicates they are a 'no_transaction' user.  \nIf there is no data at all for the user, I suppose it is best to predict them separately, according to basic statistical analysis like the one I did above; however, I will have to put in some thought into what the submitted probability will be!  \nI hope this helps!",
    "239156": "Thank you for your answer, it's good to know I wasn't wrong about those users and there is indeed no data attached to them. Hopefully when admins release new test data, we can make some inferences with these users :)"
  },
  "source": "meta"
}