{
  "id": 67401,
  "title": "How to treat totals.transactionRevenue",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/67401",
  "author_name": "PavanSangha",
  "post_date": "2018-10-02T10:54:13.567000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Clearly in the training data we have multiple users and so it is important for us to sum totals.transactionRevenue per user.  However are we then meant to obtain a separate training set with one entry per user corresponding to their total transaction revenue?  If so, how do we aggregate some of the features like geographic information? </p>\n\n<p>If we are not doing this and we stick with the same training set with duplicate entries per user, is our predictive value just their total transaction revenue repeated or do we effectively treat each row as a separate entry? </p>\n\n<p>So for example if i had a data set like this \nin the form {user, feature1, transaction value}\nfeature1 could just be some feature to do with country of order placed for example.\nrow 1={user 1 ,  'Norway' , 100} \\\nrow 2={user 1, 'England',  200}  \\\nrow 3={user 2, 'England', 100} \\</p>\n\n<p>Then my total transaction for User 1 is 300 and 100 for User 2</p>\n\n<p>Would my training set look something like this \nwith data in the form </p>\n\n<p>in the form {user, feature1, transaction value, total transaction value}\nrow 1={user 1 ,  'Norway' , 100, 300} \\\nrow 2={user 1, 'England', 200, 300} \\\nrow 3={user 2, 'England', 100, 300} \\</p>\n\n<p>Or do i get a training set with only one entry per User and pretend as if each row is unique?</p>",
  "messages": [
    {
      "id": 397333,
      "postDate": "2018-10-02T10:54:13.567Z",
      "content": "<p>Clearly in the training data we have multiple users and so it is important for us to sum totals.transactionRevenue per user.  However are we then meant to obtain a separate training set with one entry per user corresponding to their total transaction revenue?  If so, how do we aggregate some of the features like geographic information? </p>\n\n<p>If we are not doing this and we stick with the same training set with duplicate entries per user, is our predictive value just their total transaction revenue repeated or do we effectively treat each row as a separate entry? </p>\n\n<p>So for example if i had a data set like this \nin the form {user, feature1, transaction value}\nfeature1 could just be some feature to do with country of order placed for example.\nrow 1={user 1 ,  'Norway' , 100} \\\nrow 2={user 1, 'England',  200}  \\\nrow 3={user 2, 'England', 100} \\</p>\n\n<p>Then my total transaction for User 1 is 300 and 100 for User 2</p>\n\n<p>Would my training set look something like this \nwith data in the form </p>\n\n<p>in the form {user, feature1, transaction value, total transaction value}\nrow 1={user 1 ,  'Norway' , 100, 300} \\\nrow 2={user 1, 'England', 200, 300} \\\nrow 3={user 2, 'England', 100, 300} \\</p>\n\n<p>Or do i get a training set with only one entry per User and pretend as if each row is unique?</p>",
      "rawMarkdown": "Clearly in the training data we have multiple users and so it is important for us to sum totals.transactionRevenue per user.  However are we then meant to obtain a separate training set with one entry per user corresponding to their total transaction revenue?  If so, how do we aggregate some of the features like geographic information? \n\nIf we are not doing this and we stick with the same training set with duplicate entries per user, is our predictive value just their total transaction revenue repeated or do we effectively treat each row as a separate entry? \n\nSo for example if i had a data set like this \nin the form {user, feature1, transaction value}\nfeature1 could just be some feature to do with country of order placed for example.\nrow 1={user 1 ,  'Norway' , 100} \\\nrow 2={user 1, 'England',  200}  \\\nrow 3={user 2, 'England', 100} \\\n\n\nThen my total transaction for User 1 is 300 and 100 for User 2\n\nWould my training set look something like this \nwith data in the form \n\nin the form {user, feature1, transaction value, total transaction value}\nrow 1={user 1 ,  'Norway' , 100, 300} \\\nrow 2={user 1, 'England', 200, 300} \\\nrow 3={user 2, 'England', 100, 300} \\\n\nOr do i get a training set with only one entry per User and pretend as if each row is unique?"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "397333": "Clearly in the training data we have multiple users and so it is important for us to sum totals.transactionRevenue per user.  However are we then meant to obtain a separate training set with one entry per user corresponding to their total transaction revenue?  If so, how do we aggregate some of the features like geographic information? \n\nIf we are not doing this and we stick with the same training set with duplicate entries per user, is our predictive value just their total transaction revenue repeated or do we effectively treat each row as a separate entry? \n\nSo for example if i had a data set like this \nin the form {user, feature1, transaction value}\nfeature1 could just be some feature to do with country of order placed for example.\nrow 1={user 1 ,  'Norway' , 100} \\\nrow 2={user 1, 'England',  200}  \\\nrow 3={user 2, 'England', 100} \\\n\n\nThen my total transaction for User 1 is 300 and 100 for User 2\n\nWould my training set look something like this \nwith data in the form \n\nin the form {user, feature1, transaction value, total transaction value}\nrow 1={user 1 ,  'Norway' , 100, 300} \\\nrow 2={user 1, 'England', 200, 300} \\\nrow 3={user 2, 'England', 100, 300} \\\n\nOr do i get a training set with only one entry per User and pretend as if each row is unique?"
  }
}