{
  "id": 43622,
  "title": "Confused aboout the train data",
  "url": "/competitions/kkbox-churn-prediction-challenge/discussion/43622",
  "author_name": "",
  "post_date": "2017-11-17T00:45:54.939996300Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi, everyone!</p>\n\n<p>I have read the train data and the churn label confused me a lot. Some of the users in train.csv shows they are churn but the same users in train_v2.csv give me not.</p>\n\n<pre><code>          msno                                     train_v1  train_v2\n //4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=          0          1\n</code></pre>\n\n<p>And his transaction records like(only find him in transaction_v1), </p>\n\n<pre><code>                 transaction_date   membership_expire_date     is_cancel\n\n260973           20160824               20160923                     0\n\n7603788          20170224               20170326                     0\n\n11612050         20160724               20160823                     0\n\n14274185         20150918               20151018                     0\n\n17098675         20161124               20161224                     0\n\n17669724         20170119               20170218                     0\n\n18095255         20160519               20160618                     0\n\n18491652         20160130               20160229                     0\n\n18767803         20160623               20160723                     0\n</code></pre>\n\n<p>I have no idea why they became churn users. <br>\nWhat is exactly the churn prediction month for train_v1 and train_v2?</p>",
  "messages": [
    {
      "id": "244803",
      "postDate": "11/17/2017 00:45:54",
      "content": "<p>Hi, everyone!</p>\n\n<p>I have read the train data and the churn label confused me a lot. Some of the users in train.csv shows they are churn but the same users in train_v2.csv give me not.</p>\n\n<pre><code>          msno                                     train_v1  train_v2\n //4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=          0          1\n</code></pre>\n\n<p>And his transaction records like(only find him in transaction_v1), </p>\n\n<pre><code>                 transaction_date   membership_expire_date     is_cancel\n\n260973           20160824               20160923                     0\n\n7603788          20170224               20170326                     0\n\n11612050         20160724               20160823                     0\n\n14274185         20150918               20151018                     0\n\n17098675         20161124               20161224                     0\n\n17669724         20170119               20170218                     0\n\n18095255         20160519               20160618                     0\n\n18491652         20160130               20160229                     0\n\n18767803         20160623               20160723                     0\n</code></pre>\n\n<p>I have no idea why they became churn users. <br>\nWhat is exactly the churn prediction month for train_v1 and train_v2?</p>",
      "rawMarkdown": "Hi, everyone!\n\nI have read the train data and the churn label confused me a lot. Some of the users in train.csv shows they are churn but the same users in train_v2.csv give me not.\n\n         \n              msno                                     train_v1  train_v2\n     //4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=          0          1\n\nAnd his transaction records like(only find him in transaction_v1), \n\n                     transaction_date   membership_expire_date     is_cancel\n    \n    260973           20160824               20160923                     0\n    \n    7603788          20170224               20170326                     0\n    \n    11612050         20160724               20160823                     0\n    \n    14274185         20150918               20151018                     0\n    \n    17098675         20161124               20161224                     0\n    \n    17669724         20170119               20170218                     0\n    \n    18095255         20160519               20160618                     0\n    \n    18491652         20160130               20160229                     0\n    \n    18767803         20160623               20160723                     0\n\nI have no idea why they became churn users.  \nWhat is exactly the churn prediction month for train_v1 and train_v2?",
      "votes": null
    },
    {
      "id": "244880",
      "postDate": "11/17/2017 04:25:07",
      "content": "<p>The trainv2 is the dataset collected till a newer date. So the users might have canceled after purchasing the service. I have just started so I can not elaborate on this, but I believe you might find something by analysing the user logs and find patterns and extract features regarding that.</p>",
      "rawMarkdown": "The trainv2 is the dataset collected till a newer date. So the users might have canceled after purchasing the service. I have just started so I can not elaborate on this, but I believe you might find something by analysing the user logs and find patterns and extract features regarding that.",
      "votes": null
    },
    {
      "id": "244931",
      "postDate": "11/17/2017 07:20:11",
      "content": "<p>A churn user is defined as a user who does not renew service within 30 days of membership expiry date. Note, late renewals also counts as churn. For more details about how the labels were generated, please refer to the code we provided in the data section.  In this case, you can interpret the result as the user did renew his/her membership in the month of March but stopped his/her membership in the month of April.</p>",
      "rawMarkdown": "A churn user is defined as a user who does not renew service within 30 days of membership expiry date. Note, late renewals also counts as churn. For more details about how the labels were generated, please refer to the code we provided in the data section.  In this case, you can interpret the result as the user did renew his/her membership in the month of March but stopped his/her membership in the month of April.",
      "votes": null
    },
    {
      "id": "245264",
      "postDate": "11/17/2017 22:17:47",
      "content": "<p>Thanks for your clarification! We are still confused about how <strong>is_churn</strong>  of train_v1 (0) and train_v2 (1) are different from each other. The last two records in transaction file show that user <strong>//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=</strong> purchased music plan on 2017-01-19, which expired on 2017-02-18. Then, this user continued the plan on 2017-02-24 (just 7 days after the previous expiration date, less than 30 days), Thus,  the user cannot be regarded as churn one in both January and February as per your posted rule.   Thus, our question is</p>\n\n<p>What do you mean by <strong>1</strong> of <strong>is_churn</strong> in train_v2 for user <strong>//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=</strong>  ?</p>",
      "rawMarkdown": "Thanks for your clarification! We are still confused about how **is_churn**  of train_v1 (0) and train_v2 (1) are different from each other. The last two records in transaction file show that user **//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=** purchased music plan on 2017-01-19, which expired on 2017-02-18. Then, this user continued the plan on 2017-02-24 (just 7 days after the previous expiration date, less than 30 days), Thus,  the user cannot be regarded as churn one in both January and February as per your posted rule.   Thus, our question is\n\nWhat do you mean by **1** of **is_churn** in train_v2 for user **//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=**  ?",
      "votes": null
    },
    {
      "id": "248560",
      "postDate": "11/26/2017 13:41:43",
      "content": "<p>Exactly, it seems so confusing.\nI think there are some problems in the provided code. <br>\n(Before reading the code yet)</p>",
      "rawMarkdown": "Exactly, it seems so confusing.\nI think there are some problems in the provided code. <br>\n(Before reading the code yet)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 244880,
      "author_name": "kr1shna",
      "author_url": "",
      "post_date": "11/17/2017 04:25:07",
      "content": "<p>The trainv2 is the dataset collected till a newer date. So the users might have canceled after purchasing the service. I have just started so I can not elaborate on this, but I believe you might find something by analysing the user logs and find patterns and extract features regarding that.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 244931,
      "author_name": "ardenkkbox",
      "author_url": "",
      "post_date": "11/17/2017 07:20:11",
      "content": "<p>A churn user is defined as a user who does not renew service within 30 days of membership expiry date. Note, late renewals also counts as churn. For more details about how the labels were generated, please refer to the code we provided in the data section.  In this case, you can interpret the result as the user did renew his/her membership in the month of March but stopped his/her membership in the month of April.</p>",
      "votes": null,
      "replies": [
        {
          "id": 245264,
          "author_name": "jiayuzhang",
          "author_url": "",
          "post_date": "11/17/2017 22:17:47",
          "content": "<p>Thanks for your clarification! We are still confused about how <strong>is_churn</strong>  of train_v1 (0) and train_v2 (1) are different from each other. The last two records in transaction file show that user <strong>//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=</strong> purchased music plan on 2017-01-19, which expired on 2017-02-18. Then, this user continued the plan on 2017-02-24 (just 7 days after the previous expiration date, less than 30 days), Thus,  the user cannot be regarded as churn one in both January and February as per your posted rule.   Thus, our question is</p>\n\n<p>What do you mean by <strong>1</strong> of <strong>is_churn</strong> in train_v2 for user <strong>//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=</strong>  ?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 248560,
          "author_name": "sundong",
          "author_url": "",
          "post_date": "11/26/2017 13:41:43",
          "content": "<p>Exactly, it seems so confusing.\nI think there are some problems in the provided code. <br>\n(Before reading the code yet)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "244803": "Hi, everyone!\n\nI have read the train data and the churn label confused me a lot. Some of the users in train.csv shows they are churn but the same users in train_v2.csv give me not.\n\n         \n              msno                                     train_v1  train_v2\n     //4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=          0          1\n\nAnd his transaction records like(only find him in transaction_v1), \n\n                     transaction_date   membership_expire_date     is_cancel\n    \n    260973           20160824               20160923                     0\n    \n    7603788          20170224               20170326                     0\n    \n    11612050         20160724               20160823                     0\n    \n    14274185         20150918               20151018                     0\n    \n    17098675         20161124               20161224                     0\n    \n    17669724         20170119               20170218                     0\n    \n    18095255         20160519               20160618                     0\n    \n    18491652         20160130               20160229                     0\n    \n    18767803         20160623               20160723                     0\n\nI have no idea why they became churn users.  \nWhat is exactly the churn prediction month for train_v1 and train_v2?",
    "244880": "The trainv2 is the dataset collected till a newer date. So the users might have canceled after purchasing the service. I have just started so I can not elaborate on this, but I believe you might find something by analysing the user logs and find patterns and extract features regarding that.",
    "244931": "A churn user is defined as a user who does not renew service within 30 days of membership expiry date. Note, late renewals also counts as churn. For more details about how the labels were generated, please refer to the code we provided in the data section.  In this case, you can interpret the result as the user did renew his/her membership in the month of March but stopped his/her membership in the month of April.",
    "245264": "Thanks for your clarification! We are still confused about how **is_churn**  of train_v1 (0) and train_v2 (1) are different from each other. The last two records in transaction file show that user **//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=** purchased music plan on 2017-01-19, which expired on 2017-02-18. Then, this user continued the plan on 2017-02-24 (just 7 days after the previous expiration date, less than 30 days), Thus,  the user cannot be regarded as churn one in both January and February as per your posted rule.   Thus, our question is\n\nWhat do you mean by **1** of **is_churn** in train_v2 for user **//4NDf0IbQmak7rFiydVhWjlTLe947EQpOoYvcriQ4M=**  ?",
    "248560": "Exactly, it seems so confusing.\nI think there are some problems in the provided code. <br>\n(Before reading the code yet)"
  },
  "source": "meta"
}