Out of curiosity, has anybody tried fitting a regular model, without leaking the target, taking into account the time gap we are observing between training and test data? I've been off this comp for a while (waiting for new data to come in) and scrolling through hundreds of LB entries of 0.000 I am struggling to understand the state of play at the moment. Any feedback would be much appreciated.
", "messages": [ { "id": "424900", "postDate": "11/20/2018 21:12:42", "content": "Out of curiosity, has anybody tried fitting a regular model, without leaking the target, taking into account the time gap we are observing between training and test data? I've been off this comp for a while (waiting for new data to come in) and scrolling through hundreds of LB entries of 0.000 I am struggling to understand the state of play at the moment. Any feedback would be much appreciated.
", "rawMarkdown": "Out of curiosity, has anybody tried fitting a regular model, without leaking the target, taking into account the time gap we are observing between training and test data? I've been off this comp for a while (waiting for new data to come in) and scrolling through hundreds of LB entries of 0.000 I am struggling to understand the state of play at the moment. Any feedback would be much appreciated.", "votes": null }, { "id": "425140", "postDate": "11/21/2018 07:39:03", "content": "Not really. Too many unanswered questions popped up lately. That made me scratch my head quite often. Like ... :
\n\nUnfortunately the new dataset feels messy and not properly thought through.
", "rawMarkdown": "Not really. Too many unanswered questions popped up lately. That made me scratch my head quite often. Like ... :\n\n - What is the correct target (transactionRevenue/ totalTransactionRevenue). \n - What about sessionQuality and transactions column?\n\nUnfortunately the new dataset feels messy and not properly thought through.", "votes": null }, { "id": "425602", "postDate": "11/21/2018 20:42:32", "content": "Most people aren't predicting the correct variable in the latest kernels and the fact that we have basically no features for the time period we are going to predict makes this competition quite strange indeed. Take a look at https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/72231 for more information.
", "rawMarkdown": "Most people aren't predicting the correct variable in the latest kernels and the fact that we have basically no features for the time period we are going to predict makes this competition quite strange indeed. Take a look at https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/72231 for more information.", "votes": null }, { "id": "425766", "postDate": "11/22/2018 04:18:03", "content": "I spent a lot of time to figure out what happen on new data set....
\nAs I know, this competition becomes future predict question but looks it's not quiet like time series data set...\nSo it's hard to predict future revenue.
\nWhy LB have a lot of 0.00 score? it's because period of LB score is same with Test set(2018/05/1~ 2018/10/15), so we can only use the test set target, then we can get 0 rmse in LB.
\nFor Private LB, we need to predict future period(2018/12/1 ~ 2019/01/31) , so I think LB's score don't have any meaning of this competition
\nYou can refer my kernel to see what happen on this competitions:\nhttps://www.kaggle.com/super13579/ga-v2-future-purchase-prediction
Hi Dan,\nNew dataset is just raw data from BigQuery\nYou can find description here:\nhttps://support.google.com/analytics/answer/3437719?hl=en
\n\nI think we need field totalTransactionRevenue.\nDo you have any insights about Session quality? From my experience, adding this field improves validation.
", "rawMarkdown": "Hi Dan,\nNew dataset is just raw data from BigQuery\nYou can find description here:\nhttps://support.google.com/analytics/answer/3437719?hl=en\n\nI think we need field totalTransactionRevenue.\nDo you have any insights about Session quality? From my experience, adding this field improves validation.", "votes": null }, { "id": "426727", "postDate": "11/23/2018 17:55:15", "content": "@Egor,
\n\nIt is indeed BigQuery from the demo account, but the BigQuery in demo account in the link you sent, i see only datasets of 2017.\nnothing live.
\n\nAm I missing something?
", "rawMarkdown": "Egor,\n\nIt is indeed BigQuery from the demo account, but the BigQuery in demo account in the link you sent, i see only datasets of 2017.\nnothing live.\n\nAm I missing something?", "votes": null } ], "comments": [ { "id": 425140, "author_name": "springmanndaniel", "author_url": "", "post_date": "11/21/2018 07:39:03", "content": "Not really. Too many unanswered questions popped up lately. That made me scratch my head quite often. Like ... :
\n\nUnfortunately the new dataset feels messy and not properly thought through.
", "votes": null, "replies": [ { "id": 426626, "author_name": "eguliy", "author_url": "", "post_date": "11/23/2018 14:59:19", "content": "Hi Dan,\nNew dataset is just raw data from BigQuery\nYou can find description here:\nhttps://support.google.com/analytics/answer/3437719?hl=en
\n\nI think we need field totalTransactionRevenue.\nDo you have any insights about Session quality? From my experience, adding this field improves validation.
", "votes": null, "replies": [] }, { "id": 426727, "author_name": "tpthegreat", "author_url": "", "post_date": "11/23/2018 17:55:15", "content": "@Egor,
\n\nIt is indeed BigQuery from the demo account, but the BigQuery in demo account in the link you sent, i see only datasets of 2017.\nnothing live.
\n\nAm I missing something?
", "votes": null, "replies": [] } ] }, { "id": 425602, "author_name": "frjo0083", "author_url": "", "post_date": "11/21/2018 20:42:32", "content": "Most people aren't predicting the correct variable in the latest kernels and the fact that we have basically no features for the time period we are going to predict makes this competition quite strange indeed. Take a look at https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/72231 for more information.
", "votes": null, "replies": [] }, { "id": 425766, "author_name": "super13579", "author_url": "", "post_date": "11/22/2018 04:18:03", "content": "I spent a lot of time to figure out what happen on new data set....
\nAs I know, this competition becomes future predict question but looks it's not quiet like time series data set...\nSo it's hard to predict future revenue.
\nWhy LB have a lot of 0.00 score? it's because period of LB score is same with Test set(2018/05/1~ 2018/10/15), so we can only use the test set target, then we can get 0 rmse in LB.
\nFor Private LB, we need to predict future period(2018/12/1 ~ 2019/01/31) , so I think LB's score don't have any meaning of this competition
\nYou can refer my kernel to see what happen on this competitions:\nhttps://www.kaggle.com/super13579/ga-v2-future-purchase-prediction