I found 7679 fullVisitorId appearing in both train and test, with some of them have non-zero transactionRevenue (533 to be exact).
\n\nFor example, fullVisitorId 1759779356678951325 have 22 rows in train and 5 rows in test. Of the 22 in train, \"visit 11\" and \"visit 12\" are curiously missing (potential feature?). Of the 5 rows in test, \"visit 27\" is missing. So this particular user has visited 31 times in total... but for those in train, there are 6 instances where transactionRevenue were not 0 (i.e. 264790000, 258300000, 574200000, 265530000, 258900000 and 1479630000)... so for our prediction/submission, are we suppose to predict the transactionRevenue for visit 26, 28, 29, 30, 31 -- and sum them with those already known in train (i.e. 3101350000 or 21.85 after taking log)?
", "messages": [ { "id": "387290", "postDate": "09/14/2018 17:19:35", "content": "I found 7679 fullVisitorId appearing in both train and test, with some of them have non-zero transactionRevenue (533 to be exact).
\n\nFor example, fullVisitorId 1759779356678951325 have 22 rows in train and 5 rows in test. Of the 22 in train, \"visit 11\" and \"visit 12\" are curiously missing (potential feature?). Of the 5 rows in test, \"visit 27\" is missing. So this particular user has visited 31 times in total... but for those in train, there are 6 instances where transactionRevenue were not 0 (i.e. 264790000, 258300000, 574200000, 265530000, 258900000 and 1479630000)... so for our prediction/submission, are we suppose to predict the transactionRevenue for visit 26, 28, 29, 30, 31 -- and sum them with those already known in train (i.e. 3101350000 or 21.85 after taking log)?
", "rawMarkdown": "I found 7679 fullVisitorId appearing in both train and test, with some of them have non-zero transactionRevenue (533 to be exact).\n\nFor example, fullVisitorId **1759779356678951325** have 22 rows in train and 5 rows in test. Of the 22 in train, \"visit 11\" and \"visit 12\" are curiously missing (potential feature?). Of the 5 rows in test, \"visit 27\" is missing. So this particular user has visited 31 times in total... but for those in train, there are 6 instances where transactionRevenue were not 0 (i.e. 264790000, 258300000, 574200000, 265530000, 258900000 and 1479630000)... so for our prediction/submission, are we suppose to predict the transactionRevenue for visit 26, 28, 29, 30, 31 -- and sum them with those already known in train (i.e. 3101350000 or 21.85 after taking log)?", "votes": null }, { "id": "387585", "postDate": "09/15/2018 08:22:49", "content": "Maybe what is intended is to predict the total revenue per client per year. And since test is shifted by a year wrt train, you do not need to sum them up
\n\nBut maybe it is better to ask in the \"welcome\" discussion, as kaggle admins monitor that one more closely
", "rawMarkdown": "Maybe what is intended is to predict the total revenue per client **per year**. And since test is shifted by a year wrt train, you do not need to sum them up\n\nBut maybe it is better to ask in the \"welcome\" discussion, as kaggle admins monitor that one more closely", "votes": null }, { "id": "387774", "postDate": "09/15/2018 16:55:01", "content": "The datasets are split by time. You are making predictions for a time period which does not exist in the training set.
", "rawMarkdown": "The datasets are split by time. You are making predictions for a time period which does not exist in the training set.", "votes": null }, { "id": "392623", "postDate": "09/24/2018 05:13:54", "content": "Probably, for submission we need to sum only predicted transactionRevenue for the entries in the test set. However, it seems like we will be able to make better predictions by also including data from the train set, where available (need to test with/without it).
", "rawMarkdown": "Probably, for submission we need to sum only predicted transactionRevenue for the entries in the test set. However, it seems like we will be able to make better predictions by also including data from the train set, where available (need to test with/without it).", "votes": null }, { "id": "392640", "postDate": "09/24/2018 06:04:37", "content": "I'm also wondering how does visitNumber field start counting. Look at the visitor 0000677695778949032 in both train and test sets.\nTrain: years 2016,2016,2017; visitNumbers 2,3,4\nTest: years 2018,2018,2018; visitNumbers 4,5,6
\n\nSo visitNumber is NOT reset each year as I thought initially. Neither is it kept unique across years. How and when is it reset then?
", "rawMarkdown": "I'm also wondering how does visitNumber field start counting. Look at the visitor 0000677695778949032 in both train and test sets.\nTrain: years 2016,2016,2017; visitNumbers 2,3,4\nTest: years 2018,2018,2018; visitNumbers 4,5,6\n\nSo visitNumber is NOT reset each year as I thought initially. Neither is it kept unique across years. How and when is it reset then?", "votes": null }, { "id": "397030", "postDate": "10/01/2018 19:24:32", "content": "Did you find anything useful info on your finding ?
", "rawMarkdown": "Did you find anything useful info on your finding ?", "votes": null } ], "comments": [ { "id": 387585, "author_name": "mlisovyi", "author_url": "", "post_date": "09/15/2018 08:22:49", "content": "Maybe what is intended is to predict the total revenue per client per year. And since test is shifted by a year wrt train, you do not need to sum them up
\n\nBut maybe it is better to ask in the \"welcome\" discussion, as kaggle admins monitor that one more closely
", "votes": null, "replies": [] }, { "id": 387774, "author_name": "philculliton", "author_url": "", "post_date": "09/15/2018 16:55:01", "content": "The datasets are split by time. You are making predictions for a time period which does not exist in the training set.
", "votes": null, "replies": [] }, { "id": 392623, "author_name": "fingoldo", "author_url": "", "post_date": "09/24/2018 05:13:54", "content": "Probably, for submission we need to sum only predicted transactionRevenue for the entries in the test set. However, it seems like we will be able to make better predictions by also including data from the train set, where available (need to test with/without it).
", "votes": null, "replies": [] }, { "id": 392640, "author_name": "fingoldo", "author_url": "", "post_date": "09/24/2018 06:04:37", "content": "I'm also wondering how does visitNumber field start counting. Look at the visitor 0000677695778949032 in both train and test sets.\nTrain: years 2016,2016,2017; visitNumbers 2,3,4\nTest: years 2018,2018,2018; visitNumbers 4,5,6
\n\nSo visitNumber is NOT reset each year as I thought initially. Neither is it kept unique across years. How and when is it reset then?
", "votes": null, "replies": [ { "id": 397030, "author_name": "", "author_url": "", "post_date": "10/01/2018 19:24:32", "content": "Did you find anything useful info on your finding ?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "387290": "I found 7679 fullVisitorId appearing in both train and test, with some of them have non-zero transactionRevenue (533 to be exact).\n\nFor example, fullVisitorId **1759779356678951325** have 22 rows in train and 5 rows in test. Of the 22 in train, \"visit 11\" and \"visit 12\" are curiously missing (potential feature?). Of the 5 rows in test, \"visit 27\" is missing. So this particular user has visited 31 times in total... but for those in train, there are 6 instances where transactionRevenue were not 0 (i.e. 264790000, 258300000, 574200000, 265530000, 258900000 and 1479630000)... so for our prediction/submission, are we suppose to predict the transactionRevenue for visit 26, 28, 29, 30, 31 -- and sum them with those already known in train (i.e. 3101350000 or 21.85 after taking log)?", "387585": "Maybe what is intended is to predict the total revenue per client **per year**. And since test is shifted by a year wrt train, you do not need to sum them up\n\nBut maybe it is better to ask in the \"welcome\" discussion, as kaggle admins monitor that one more closely", "387774": "The datasets are split by time. You are making predictions for a time period which does not exist in the training set.", "392623": "Probably, for submission we need to sum only predicted transactionRevenue for the entries in the test set. However, it seems like we will be able to make better predictions by also including data from the train set, where available (need to test with/without it).", "392640": "I'm also wondering how does visitNumber field start counting. Look at the visitor 0000677695778949032 in both train and test sets.\nTrain: years 2016,2016,2017; visitNumbers 2,3,4\nTest: years 2018,2018,2018; visitNumbers 4,5,6\n\nSo visitNumber is NOT reset each year as I thought initially. Neither is it kept unique across years. How and when is it reset then?", "397030": "Did you find anything useful info on your finding ?" }, "source": "meta" }