{ "id": 68235, "title": "Insight or Dodgy", "url": "/competitions/ga-customer-revenue-prediction/discussion/68235", "author_name": "Scirpus", "post_date": "2018-10-10T16:55:14.079000", "votes": 49, "comment_count": 63, "views": 0, "content": "

Anyone else worried how we use the demo data a.ka https://analytics.google.com/analytics/web/demoAccount

\n\n

We know things like revenue per month etc so we are asked to predict things that we can almost look up!

", "messages": [ { "id": 401880, "postDate": "2018-10-10T20:28:51.997Z", "content": "

I stopped thinking about this competition for a few days when I realized that most of the revenues was quite easily recoverable from the demo... :(\nMy session level cv is 0.55 after integrating an extraction from the demo. \nAs I have already mentioned here https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/68164 this morning, the days of the competition are unfortunately numbered unless Kaggle finds a solution for this data leak.

", "rawMarkdown": "I stopped thinking about this competition for a few days when I realized that most of the revenues was quite easily recoverable from the demo... :(\nMy session level cv is 0.55 after integrating an extraction from the demo. \nAs I have already mentioned here [https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/68164][1] this morning, the days of the competition are unfortunately numbered unless Kaggle finds a solution for this data leak.\n\n\n [1]: https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/68164", "votes": 90, "replies": [ { "id": 401882, "postDate": "2018-10-10T20:42:16.907Z", "content": "

Sorry if my question looks silly but ff this is a leak and your CV is 0.55, how come you're still at rank 545 ?

", "rawMarkdown": "Sorry if my question looks silly but ff this is a leak and your CV is 0.55, how come you're still at rank 545 ? ", "votes": 1 }, { "id": 401886, "postDate": "2018-10-10T20:48:49.113Z", "content": "

I didn't submit it. I do not want to be accused of condemning the competition while many have worked a lot on it :))

", "rawMarkdown": "I didn't submit it. I do not want to be accused of condemning the competition while many have worked a lot on it :))\n", "votes": 50 }, { "id": 401889, "postDate": "2018-10-10T20:53:34.573Z", "content": "

This is intresting!!! i want to see how Kaggle will deal with high LB scores that contain such an information. I mean that someone can use a part of these data and get a cv of 1.45 for example. This kind of CV will receive a gold medal....

", "rawMarkdown": "This is intresting!!! i want to see how Kaggle will deal with high LB scores that contain such an information. I mean that someone can use a part of these data and get a cv of 1.45 for example. This kind of CV will receive a gold medal....", "votes": 1 }, { "id": 401894, "postDate": "2018-10-10T20:59:12.093Z", "content": "

Jerome, can you please submit it? so that we understand that the leak is also applicable to the test set.

", "rawMarkdown": "Jerome, can you please submit it? so that we understand that the leak is also applicable to the test set.", "votes": 3 }, { "id": 401898, "postDate": "2018-10-10T21:13:53.157Z", "content": "

+1 for submission. If such leak indeed exists, low LB score would a) force Kaggle to do something about it b) make more people aware of it + help to avoid wasting even more of their time with this competition.

", "rawMarkdown": "+1 for submission. If such leak indeed exists, low LB score would a) force Kaggle to do something about it b) make more people aware of it + help to avoid wasting even more of their time with this competition.", "votes": 12 }, { "id": 401901, "postDate": "2018-10-10T21:27:03.180Z", "content": "

Please don't kill me! This submission is just to show the leak :(

", "rawMarkdown": "Please don't kill me! This submission is just to show the leak :(", "votes": 75 }, { "id": 401905, "postDate": "2018-10-10T21:35:47.887Z", "content": "

There there :( :(

\n\n

Why did they allow to use external data then ?

", "rawMarkdown": "There there :( :(\n\nWhy did they allow to use external data then ?\n ", "votes": 1 }, { "id": 401906, "postDate": "2018-10-10T21:36:27.733Z", "content": "

I was working on the leak today, looks like I won't have to do the work after all. Hopefully you did it all in R! :)

", "rawMarkdown": "I was working on the leak today, looks like I won't have to do the work after all. Hopefully you did it all in R! :)", "votes": 1 }, { "id": 401908, "postDate": "2018-10-10T21:41:48.143Z", "content": "

That is the question @Indranil... The link to demoAccount is available for 11 days and no reaction.

", "rawMarkdown": "That is the question @Indranil... The link to demoAccount is available for 11 days and no reaction.", "votes": 11 }, { "id": 401910, "postDate": "2018-10-10T21:42:27.127Z", "content": "

Thanks Jerome...BTW, so much of our time wasted :(

", "rawMarkdown": "Thanks Jerome...BTW, so much of our time wasted :(", "votes": 7 }, { "id": 401912, "postDate": "2018-10-10T21:45:09.890Z", "content": "

And there it is...nice going Jerome!

", "rawMarkdown": "And there it is...nice going Jerome!", "votes": 1 }, { "id": 401914, "postDate": "2018-10-10T21:47:12.157Z", "content": "

Oh my god....Now i believe that these data should become available for all by the competition host. Otherwise this competition has not any meaning at all. Eitherway, there was not business value from the first place due to the poor metric, as mentioned by others.... I cannot believe that we lost so much valuable free time for nothing

", "rawMarkdown": "Oh my god....Now i believe that these data should become available for all by the competition host. Otherwise this competition has not any meaning at all. Eitherway, there was not business value from the first place due to the poor metric, as mentioned by others.... I cannot believe that we lost so much valuable free time for nothing", "votes": 4 }, { "id": 401916, "postDate": "2018-10-10T21:50:34.737Z", "content": "

I learned a lot so this isn't wasted time; but I was hoping to start building a portfolio with this competition. Hopefully another interesting csv data competition comes up soon!

", "rawMarkdown": "I learned a lot so this isn't wasted time; but I was hoping to start building a portfolio with this competition. Hopefully another interesting csv data competition comes up soon!", "votes": 9 }, { "id": 401922, "postDate": "2018-10-10T22:01:00.703Z", "content": "

I disagree about the learning thing. You can learn only when you have done your best, you pushed yourself to find the best features, models etc. and finally look at the winner's solutions, in order to compare what you did in a wrong way and what you did correct. Now we miss that part.

", "rawMarkdown": "I disagree about the learning thing. You can learn only when you have done your best, you pushed yourself to find the best features, models etc. and finally look at the winner's solutions, in order to compare what you did in a wrong way and what you did correct. Now we miss that part.", "votes": 7 }, { "id": 401927, "postDate": "2018-10-10T22:04:27.847Z", "content": "

Totally agree with you. I had a list of ideas I wanted to try, and now it feels a bit pointless.

", "rawMarkdown": "Totally agree with you. I had a list of ideas I wanted to try, and now it feels a bit pointless. ", "votes": 4 }, { "id": 401950, "postDate": "2018-10-10T22:59:50.737Z", "content": "

Thanks for the submission, upvoted. It's better to find it now than at the end of the competition...

\n\n

I don't see how Kaggle can fix this. Closing external data is difficult since it is not a kernel only competition and using the leak makes this competition a joke...

", "rawMarkdown": "Thanks for the submission, upvoted. It's better to find it now than at the end of the competition... \n\nI don't see how Kaggle can fix this. Closing external data is difficult since it is not a kernel only competition and using the leak makes this competition a joke...", "votes": 2 }, { "id": 401992, "postDate": "2018-10-11T02:31:28.990Z", "content": "

I decided to stop thinking about this competition. I'm very disappointed now. \nI've learned a lot of things during working on this competition. This is always my prize. \nThanks for exposure of the leak. You're saving the incredible amount of hours of kaggler!

", "rawMarkdown": "I decided to stop thinking about this competition. I'm very disappointed now. \nI've learned a lot of things during working on this competition. This is always my prize. \nThanks for exposure of the leak. You're saving the incredible amount of hours of kaggler!", "votes": 12 }, { "id": 402007, "postDate": "2018-10-11T03:11:51.870Z", "content": "

I don't know why but I had a feeling it would end like this. Particularly after they announced that external data was available.

", "rawMarkdown": "I don't know why but I had a feeling it would end like this. Particularly after they announced that external data was available.", "votes": 1 }, { "id": 402016, "postDate": "2018-10-11T03:38:23.080Z", "content": "

Thanks for this. Really saved everyone a lot of time.

", "rawMarkdown": "Thanks for this. Really saved everyone a lot of time." }, { "id": 402053, "postDate": "2018-10-11T05:45:33.740Z", "content": "

Thanks Jerome for your investigations ! This competition is a farce from start to finish....

", "rawMarkdown": "Thanks Jerome for your investigations ! This competition is a farce from start to finish....", "votes": 11 }, { "id": 402067, "postDate": "2018-10-11T06:12:58.413Z", "content": "

Thanks Jerome for submitting!

", "rawMarkdown": "Thanks Jerome for submitting!", "votes": 1 }, { "id": 402069, "postDate": "2018-10-11T06:19:21.527Z", "content": "

@Jerome

\n\n

Thank you for submitting this \"joke\" !

", "rawMarkdown": "@Jerome\n\nThank you for submitting this \"joke\" ! ", "votes": 1 }, { "id": 402071, "postDate": "2018-10-11T06:25:12.593Z", "content": "

Thanks for sharing Jerome!

", "rawMarkdown": "Thanks for sharing Jerome!", "votes": 1 }, { "id": 402182, "postDate": "2018-10-11T09:45:47.630Z", "content": "

> Jerome Vallet wrote\n> \n> > I didn't submit it. I do not want to be accused of condemning the competition while many have worked a lot on it :))\n>

\n\n

Anyway, @emorej you did it;) The temptation is great in such cases! But thank you for sharing your insight and your honesty before community. I still have not started the competition as if I felt something after Santander did. So, now rule of the last kaggle is \"Before you start the competition data must brew\"!

", "rawMarkdown": "\n> **Jerome Vallet wrote**\n> \n> > I didn't submit it. I do not want to be accused of condemning the competition while many have worked a lot on it :))\n> \n\nAnyway, @emorej you did it;) The temptation is great in such cases! But thank you for sharing your insight and your honesty before community. I still have not started the competition as if I felt something after Santander did. So, now rule of the *last* kaggle is *\"Before you start the competition data must brew\"*!" }, { "id": 402327, "postDate": "2018-10-11T14:31:20.787Z", "content": "

Sergey, Jérôme did not do it by greed but simply because lots of kagglers (including me) asked him to check if his local CV would translate to LB.

\n\n

This also helps trying to get Kaggle to react on that since the rules don't excluded using Demo Account data...

", "rawMarkdown": "Sergey, Jérôme did not do it by greed but simply because lots of kagglers (including me) asked him to check if his local CV would translate to LB.\n\nThis also helps trying to get Kaggle to react on that since the rules don't excluded using Demo Account data...", "votes": 3 }, { "id": 402351, "postDate": "2018-10-11T14:54:53.787Z", "content": "

Olivier, Maybe I missed something in this thread. So, my apologize. The most important thing is that the leak was opened early and there is time to solve this problem. But it remains an open question for me: \"Why Jerome and not somebody from Kaggle Team?\"

", "rawMarkdown": "Olivier, Maybe I missed something in this thread. So, my apologize. The most important thing is that the *leak* was opened early and there is time to solve this problem. But it remains an open question for me: *\"Why Jerome and not somebody from Kaggle Team?\"*", "votes": 2 }, { "id": 402354, "postDate": "2018-10-11T14:59:01.530Z", "content": "

Good question Sergey ;-)

", "rawMarkdown": "Good question Sergey ;-) ", "votes": 1 }, { "id": 402382, "postDate": "2018-10-11T16:00:58.240Z", "content": "

thanks jerome, you saved a lot of time for all of us. I am a beginner and this is my first competition so it is really sad that it turned out this way.

\n\n

olivier thanks as well, i learned a lot from your kernels.

\n\n

but out of curiosity i matched the visitid with the analytics client id and that alone increased the cv and lb score by a lot.

", "rawMarkdown": "thanks jerome, you saved a lot of time for all of us. I am a beginner and this is my first competition so it is really sad that it turned out this way.\n\nolivier thanks as well, i learned a lot from your kernels.\n\nbut out of curiosity i matched the visitid with the analytics client id and that alone increased the cv and lb score by a lot.", "votes": 3 }, { "id": 402426, "postDate": "2018-10-11T17:43:40.207Z", "content": "

Thanks Jerome -- at this rate perhaps I shouldn't even bother starting any competition until at least a month into it :-(

", "rawMarkdown": "Thanks Jerome -- at this rate perhaps I shouldn't even bother starting any competition until at least a month into it :-(", "votes": 3 }, { "id": 403089, "postDate": "2018-10-12T20:56:57.750Z", "content": "

Same! I was pumped to get my hands dirty on this one.

", "rawMarkdown": "Same! I was pumped to get my hands dirty on this one.", "votes": 1 }, { "id": 404469, "postDate": "2018-10-15T18:45:54.137Z", "content": "

Thanks Jerome for exposing the leak. You saved a lot of time

", "rawMarkdown": "Thanks Jerome for exposing the leak. You saved a lot of time" }, { "id": 408635, "postDate": "2018-10-23T07:41:44.960Z", "content": "

I am new to this.. but can anyone tell me how is the revenue in the leaked data and totals.transactionRevenue related?

", "rawMarkdown": "I am new to this.. but can anyone tell me how is the revenue in the leaked data and totals.transactionRevenue related?" }, { "id": 417602, "postDate": "2018-11-08T13:56:39.150Z", "content": "

what score did you get without the leak?

", "rawMarkdown": "what score did you get without the leak? " } ] }, { "id": 401759, "postDate": "2018-10-10T16:55:14.080Z", "content": "

Anyone else worried how we use the demo data a.ka https://analytics.google.com/analytics/web/demoAccount

\n\n

We know things like revenue per month etc so we are asked to predict things that we can almost look up!

", "rawMarkdown": "Anyone else worried how we use the demo data a.ka https://analytics.google.com/analytics/web/demoAccount\n\nWe know things like revenue per month etc so we are asked to predict things that we can almost look up!\n\n", "votes": 48 }, { "id": 401799, "postDate": "2018-10-10T18:07:03.143Z", "content": "

Maybe the point of this competition is that we all become more familiar with Google Analytics...

\n\n

@ogrellier - I have an analytics account I haven't looked at for a while. I was able to sign in to that and then access the demo account by by pressing the \"Access Demo Account\" button in the link Scirpus shared.

", "rawMarkdown": "Maybe the point of this competition is that we all become more familiar with Google Analytics...\n\n@ogrellier - I have an analytics account I haven't looked at for a while. I was able to sign in to that and then access the demo account by by pressing the \"Access Demo Account\" button in the link Scirpus shared. ", "votes": 13 }, { "id": 401915, "postDate": "2018-10-10T21:48:07.250Z", "content": "

I'll share my findings about the demo data... Jerome can confirm if I have the right insights.

\n\n

Go to User Explorer, enter the dates for the test data.\nClient Ids look like 1092690470.1510359633\nThe second part of the ID (e.g. 1510359633) corresponds to the time stamp (or visitID) of the first session of the visitor. That's how I was planning to link it to the test+train data. Sometimes you might not have the first visit in the data.

\n\n

If you click on a clientID, you will see the details of the sessions. There's a bigquery ID in there, but it doesn't seem to match with our fullVisitorID.

\n\n

Seem like you can download the rows 5000 at a time. I'm sure Jerome found a better way. Possibly you can filter by non bounce sessions.

\n\n

Initially I tried to only get the data with >0 sales. However I ran into visitors with missing sessions in the test data. I think that our test data is a weird subset of the original one. I even saw a visitor for which all but one of the sessions were removed.

\n\n

So, you might have more success instead if you focus on visitors with 0 sales. If sessions are removed from those, you still get sales=0...

", "rawMarkdown": "I'll share my findings about the demo data... Jerome can confirm if I have the right insights.\n\nGo to User Explorer, enter the dates for the test data.\nClient Ids look like 1092690470.1510359633\nThe second part of the ID (e.g. 1510359633) corresponds to the time stamp (or visitID) of the first session of the visitor. That's how I was planning to link it to the test+train data. Sometimes you might not have the first visit in the data.\n\nIf you click on a clientID, you will see the details of the sessions. There's a bigquery ID in there, but it doesn't seem to match with our fullVisitorID.\n\nSeem like you can download the rows 5000 at a time. I'm sure Jerome found a better way. Possibly you can filter by non bounce sessions.\n\nInitially I tried to only get the data with >0 sales. However I ran into visitors with missing sessions in the test data. I think that our test data is a weird subset of the original one. I even saw a visitor for which all but one of the sessions were removed.\n\nSo, you might have more success instead if you focus on visitors with 0 sales. If sessions are removed from those, you still get sales=0...\n\n", "votes": 10, "replies": [ { "id": 402142, "postDate": "2018-10-11T08:30:21.353Z", "content": "

I did fast demo account overview, so from what I found, the procedure to get exact revenue should look like this:

\n\n
    \n
  1. On User explorer select test data timeframe, order by revenue, save user ids of paying users. Number of such users is less than 10_000, so you will need just two export each by 5_000.
    2. \n
  2. Click to each user to get User report and export the data, data has all details activities of the user.
    3. \n
  3. To map competition users and ga demo data user, use pageviews for the session, they are one to one, competition hits are sum of events and ecommerce events, and session start time which as I see is the same at least to minutes level.
    4. \n
  4. to get revenue find last purchase within a session (there can be several purchases), that is a 'totals.transactionRevenue' from competition data.
    5. \n
", "rawMarkdown": "I did fast demo account overview, so from what I found, the procedure to get exact revenue should look like this:\n\n1. On User explorer select test data timeframe, order by revenue, save user ids of paying users. Number of such users is less than 10_000, so you will need just two export each by 5_000.\n2. Click to each user to get User report and export the data, data has all details activities of the user.\n3. To map competition users and ga demo data user, use pageviews for the session, they are one to one, competition hits are sum of events and ecommerce events, and session start time which as I see is the same at least to minutes level.\n4. to get revenue find last purchase within a session (there can be several purchases), that is a 'totals.transactionRevenue' from competition data.\n", "votes": 13 }, { "id": 403105, "postDate": "2018-10-12T21:57:02.840Z", "content": "

For step 2, did you find a way to do that automatically for all 10 000 users ? Thanks for the tuto !

", "rawMarkdown": "For step 2, did you find a way to do that automatically for all 10 000 users ? Thanks for the tuto !" }, { "id": 403356, "postDate": "2018-10-13T12:58:32.227Z", "content": "

Antoine, I did not try to automate this process. It is logical to use Analytics Reporting API instead of Google Analytics UI. Unfortunately Reporting API is disabled for demo account.

", "rawMarkdown": "Antoine, I did not try to automate this process. It is logical to use Analytics Reporting API instead of Google Analytics UI. Unfortunately Reporting API is disabled for demo account." } ] }, { "id": 401955, "postDate": "2018-10-10T23:42:12.773Z", "content": "

this competition is getting better (or worse) every day between the CV and now the leak... Another Santander (most recent one) competition style ?

", "rawMarkdown": "this competition is getting better (or worse) every day between the CV and now the leak... Another Santander (most recent one) competition style ?", "votes": 3, "replies": [ { "id": 401957, "postDate": "2018-10-10T23:51:23.450Z", "content": "

There might be a couple data points left once this leak is fully exploited...

", "rawMarkdown": "There might be a couple data points left once this leak is fully exploited..." } ] }, { "id": 401781, "postDate": "2018-10-10T17:44:26.597Z", "content": "

don't know, I already have an analytics account and it seems I can't open the demo account due to that !\nWe'll watch your progress on the LB ;-)

", "rawMarkdown": "don't know, I already have an analytics account and it seems I can't open the demo account due to that !\nWe'll watch your progress on the LB ;-)", "votes": 4, "replies": [ { "id": 401790, "postDate": "2018-10-10T17:55:43.677Z", "content": "

There does seem to be enough LB movement that I thought you released another kernel ;)

", "rawMarkdown": "There does seem to be enough LB movement that I thought you released another kernel ;)", "votes": 7 }, { "id": 401829, "postDate": "2018-10-10T18:51:55.090Z", "content": "

You are right, we are all waiting for another amazing Kernel from Olivier.

\n\n

I suspect some lazyness due to his comfortable rank ;-)

", "rawMarkdown": "You are right, we are all waiting for another amazing Kernel from Olivier. \n\nI suspect some lazyness due to his comfortable rank ;-)", "votes": 2 }, { "id": 401871, "postDate": "2018-10-10T20:14:29.317Z", "content": "

Indeed, the forward sessions variables seem to boost the LB. But what worries me is that in my sense these kind of features are not appropriate. In a time frame of t we do not have information of t+1. Hence, when predicting a transaction value of the latest test set rows, we do not know if a visitor has visits that are not present on the test set. Thats means, that this kind of model has incomplete data to predict from...\nBy using only lag features i managed to obtain a CV of 1.538 but the LB is 1.4242. Olivier's kernel with higher CV obtains a much better LB score.

\n\n

This competition is a real puzzle !!! :)

", "rawMarkdown": "Indeed, the forward sessions variables seem to boost the LB. But what worries me is that in my sense these kind of features are not appropriate. In a time frame of t we do not have information of t+1. Hence, when predicting a transaction value of the latest test set rows, we do not know if a visitor has visits that are not present on the test set. Thats means, that this kind of model has incomplete data to predict from...\nBy using only lag features i managed to obtain a CV of 1.538 but the LB is 1.4242. Olivier's kernel with higher CV obtains a much better LB score.\n\nThis competition is a real puzzle !!! :)", "votes": 4 }, { "id": 402066, "postDate": "2018-10-11T06:10:55.850Z", "content": "

@Dimosthenis Karaflos

\n\n

\"In a time frame of t we do not have information of t+1\"... So right !

", "rawMarkdown": "@Dimosthenis Karaflos\n\n\"*In a time frame of t we do not have information of t+1*\"... So right ! ", "votes": 1 }, { "id": 402183, "postDate": "2018-10-11T09:47:17.663Z", "content": "

From my understanding, when predicting a transaction value of the latest test set rows, , the forward session variable is likely to be NaT since we don't know whether there will be another session in the future, which is fine as it doesn't provide useful information. When it comes to other test set rows, we know the chance of purchasing is relatively small if another session of the visitor can be observed later. That information is quite valuable and should help to make better prediction. So from my perspective these feature engineering seems okay.

", "rawMarkdown": "From my understanding, when predicting a transaction value of the latest test set rows, , the forward session variable is likely to be NaT since we don't know whether there will be another session in the future, which is fine as it doesn't provide useful information. When it comes to other test set rows, we know the chance of purchasing is relatively small if another session of the visitor can be observed later. That information is quite valuable and should help to make better prediction. So from my perspective these feature engineering seems okay." } ] }, { "id": 401936, "postDate": "2018-10-10T22:19:12.887Z", "content": "

https://datastudio.google.com/u/0/reporting/0B2-rNcnRS4x5UG50LTBMT0E4aXM/page/nQN

\n\n

I discovered this when I tried to confirm the contents of the demo account. In Google DataStudio we can see the Google Merchandise Store GA report. I think that we can also label this with a correct answer. I may be wrong because I have not looked into this in detail. I'm seeking data analysis competition...

", "rawMarkdown": "https://datastudio.google.com/u/0/reporting/0B2-rNcnRS4x5UG50LTBMT0E4aXM/page/nQN\n\nI discovered this when I tried to confirm the contents of the demo account. In Google DataStudio we can see the Google Merchandise Store GA report. I think that we can also label this with a correct answer. I may be wrong because I have not looked into this in detail. I'm seeking data analysis competition...", "votes": 2, "replies": [ { "id": 401937, "postDate": "2018-10-10T22:20:22.493Z", "content": "

It seems that we can also download the report by specifying the period.

", "rawMarkdown": "It seems that we can also download the report by specifying the period." } ] }, { "id": 401872, "postDate": "2018-10-10T20:17:50.993Z", "content": "

I was at first. But then I spent a good day and a half trying to see if there was anything useful from demo Account, but I haven't been able to find anything.

\n\n

The data from the demo Account doesn't match well with the data in the competition. For instance, the total revenue in the demo account for the training period is about 4 times the amount of revenue in the competition.

\n\n

Edit: I guess it turns out that there is a way to use the demo data. See Jerome's post for more detail.

", "rawMarkdown": "I was at first. But then I spent a good day and a half trying to see if there was anything useful from demo Account, but I haven't been able to find anything.\n\nThe data from the demo Account doesn't match well with the data in the competition. For instance, the total revenue in the demo account for the training period is about 4 times the amount of revenue in the competition.\n\nEdit: I guess it turns out that there is a way to use the demo data. See Jerome's post for more detail.\n", "votes": 2 }, { "id": 402001, "postDate": "2018-10-11T03:00:28.077Z", "content": "

May be the organizers should stop this competition as it stands now. i.e. prior to leak. So that efforts put in could be saved.

", "rawMarkdown": "May be the organizers should stop this competition as it stands now. i.e. prior to leak. So that efforts put in could be saved.", "votes": -5, "replies": [ { "id": 402002, "postDate": "2018-10-11T03:03:29.393Z", "rawMarkdown": "", "isDeleted": true }, { "id": 402005, "postDate": "2018-10-11T03:09:11.970Z", "content": "

I don't think it would be a fair solution as well. Some people efforts may not be reflected in the current state of the LB-PB and there was still 1 month remaining. In addition, a lot of the current medals would be given to people just submitting recent high scoring kernels.

", "rawMarkdown": "I don't think it would be a fair solution as well. Some people efforts may not be reflected in the current state of the LB-PB and there was still 1 month remaining. In addition, a lot of the current medals would be given to people just submitting recent high scoring kernels. ", "votes": 5 }, { "id": 402029, "postDate": "2018-10-11T04:05:58.830Z", "content": "

That wouldn't be fair - many people are testing locally, and budgeting their time around work/life schedules knowing that the competitions was supposed to be going on for another month.

", "rawMarkdown": "That wouldn't be fair - many people are testing locally, and budgeting their time around work/life schedules knowing that the competitions was supposed to be going on for another month.", "votes": 3 }, { "id": 402033, "postDate": "2018-10-11T04:14:11.970Z", "content": "

Okay agree, that was one of the option. Other options are to exploit leak for the remaining time, or kaggle does something to fix data.

", "rawMarkdown": "Okay agree, that was one of the option. Other options are to exploit leak for the remaining time, or kaggle does something to fix data. ", "votes": 1 } ] }, { "id": 402295, "postDate": "2018-10-11T13:23:59.257Z", "content": "

Kaggle's turning into a sieve with these leak incidents in the recent months ;)

", "rawMarkdown": "Kaggle's turning into a sieve with these leak incidents in the recent months ;)", "replies": [ { "id": 402330, "postDate": "2018-10-11T14:33:36.110Z", "content": "

How are those incidents dealt with in the past ?

", "rawMarkdown": "How are those incidents dealt with in the past ?", "votes": 2 }, { "id": 402347, "postDate": "2018-10-11T14:51:27.080Z", "content": "

@Regan most recent leakage was for Airbus but I believe Santander was a data property and not a leak :)

", "rawMarkdown": "@Regan most recent leakage was for Airbus but I believe Santander was a data property and not a leak :)", "votes": 1 } ] }, { "id": 402292, "postDate": "2018-10-11T13:19:04.930Z", "content": "

Oh my, this are sad news.

", "rawMarkdown": "Oh my, this are sad news." }, { "id": 409878, "postDate": "2018-10-25T02:02:21.953Z", "rawMarkdown": "", "isDeleted": true }, { "id": 408072, "postDate": "2018-10-22T09:12:47.493Z", "rawMarkdown": "", "isDeleted": true }, { "id": 405110, "postDate": "2018-10-16T21:58:03.050Z", "rawMarkdown": "", "isDeleted": true }, { "id": 401911, "postDate": "2018-10-10T21:42:47.343Z", "rawMarkdown": "", "votes": 2, "isDeleted": true } ], "comments": [ { "id": 401880, "author_name": "Jerome Vallet", "author_url": "", "post_date": "2018-10-10T20:28:51.997000", "content": "

I stopped thinking about this competition for a few days when I realized that most of the revenues was quite easily recoverable from the demo... :(\nMy session level cv is 0.55 after integrating an extraction from the demo. \nAs I have already mentioned here https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/68164 this morning, the days of the competition are unfortunately numbered unless Kaggle finds a solution for this data leak.

", "votes": 90, "replies": [ { "id": 401882, "author_name": "olivier", "author_url": "", "post_date": "2018-10-10T20:42:16.907000", "content": "

Sorry if my question looks silly but ff this is a leak and your CV is 0.55, how come you're still at rank 545 ?

", "votes": 1, "replies": [] }, { "id": 401886, "author_name": "Jerome Vallet", "author_url": "", "post_date": "2018-10-10T20:48:49.113000", "content": "

I didn't submit it. I do not want to be accused of condemning the competition while many have worked a lot on it :))

", "votes": 50, "replies": [] }, { "id": 401889, "author_name": "Dimosthenis Karaflos", "author_url": "", "post_date": "2018-10-10T20:53:34.573000", "content": "

This is intresting!!! i want to see how Kaggle will deal with high LB scores that contain such an information. I mean that someone can use a part of these data and get a cv of 1.45 for example. This kind of CV will receive a gold medal....

", "votes": 1, "replies": [] }, { "id": 401894, "author_name": "Ahmet Erdem", "author_url": "", "post_date": "2018-10-10T20:59:12.093000", "content": "

Jerome, can you please submit it? so that we understand that the leak is also applicable to the test set.

", "votes": 3, "replies": [] }, { "id": 401898, "author_name": "Herra Huu", "author_url": "", "post_date": "2018-10-10T21:13:53.157000", "content": "

+1 for submission. If such leak indeed exists, low LB score would a) force Kaggle to do something about it b) make more people aware of it + help to avoid wasting even more of their time with this competition.

", "votes": 12, "replies": [] }, { "id": 401901, "author_name": "Jerome Vallet", "author_url": "", "post_date": "2018-10-10T21:27:03.180000", "content": "

Please don't kill me! This submission is just to show the leak :(

", "votes": 75, "replies": [] }, { "id": 401905, "author_name": "Indranil Bhattacharya", "author_url": "", "post_date": "2018-10-10T21:35:47.887000", "content": "

There there :( :(

\n\n

Why did they allow to use external data then ?

", "votes": 1, "replies": [] }, { "id": 401906, "author_name": "S D", "author_url": "", "post_date": "2018-10-10T21:36:27.733000", "content": "

I was working on the leak today, looks like I won't have to do the work after all. Hopefully you did it all in R! :)

", "votes": 1, "replies": [] }, { "id": 401908, "author_name": "Jerome Vallet", "author_url": "", "post_date": "2018-10-10T21:41:48.143000", "content": "

That is the question @Indranil... The link to demoAccount is available for 11 days and no reaction.

", "votes": 11, "replies": [] }, { "id": 401910, "author_name": "Kain", "author_url": "", "post_date": "2018-10-10T21:42:27.127000", "content": "

Thanks Jerome...BTW, so much of our time wasted :(

", "votes": 7, "replies": [] }, { "id": 401912, "author_name": "JohnM", "author_url": "", "post_date": "2018-10-10T21:45:09.890000", "content": "

And there it is...nice going Jerome!

", "votes": 1, "replies": [] }, { "id": 401914, "author_name": "Dimosthenis Karaflos", "author_url": "", "post_date": "2018-10-10T21:47:12.157000", "content": "

Oh my god....Now i believe that these data should become available for all by the competition host. Otherwise this competition has not any meaning at all. Eitherway, there was not business value from the first place due to the poor metric, as mentioned by others.... I cannot believe that we lost so much valuable free time for nothing

", "votes": 4, "replies": [] }, { "id": 401916, "author_name": "S D", "author_url": "", "post_date": "2018-10-10T21:50:34.737000", "content": "

I learned a lot so this isn't wasted time; but I was hoping to start building a portfolio with this competition. Hopefully another interesting csv data competition comes up soon!

", "votes": 9, "replies": [] }, { "id": 401922, "author_name": "Dimosthenis Karaflos", "author_url": "", "post_date": "2018-10-10T22:01:00.703000", "content": "

I disagree about the learning thing. You can learn only when you have done your best, you pushed yourself to find the best features, models etc. and finally look at the winner's solutions, in order to compare what you did in a wrong way and what you did correct. Now we miss that part.

", "votes": 7, "replies": [] }, { "id": 401927, "author_name": "S D", "author_url": "", "post_date": "2018-10-10T22:04:27.847000", "content": "

Totally agree with you. I had a list of ideas I wanted to try, and now it feels a bit pointless.

", "votes": 4, "replies": [] }, { "id": 401950, "author_name": "Aguiar", "author_url": "", "post_date": "2018-10-10T22:59:50.737000", "content": "

Thanks for the submission, upvoted. It's better to find it now than at the end of the competition...

\n\n

I don't see how Kaggle can fix this. Closing external data is difficult since it is not a kernel only competition and using the leak makes this competition a joke...

", "votes": 2, "replies": [] }, { "id": 401992, "author_name": "Youhan Lee", "author_url": "", "post_date": "2018-10-11T02:31:28.990000", "content": "

I decided to stop thinking about this competition. I'm very disappointed now. \nI've learned a lot of things during working on this competition. This is always my prize. \nThanks for exposure of the leak. You're saving the incredible amount of hours of kaggler!

", "votes": 12, "replies": [] }, { "id": 402007, "author_name": "Adri", "author_url": "", "post_date": "2018-10-11T03:11:51.870000", "content": "

I don't know why but I had a feeling it would end like this. Particularly after they announced that external data was available.

", "votes": 1, "replies": [] }, { "id": 402016, "author_name": "Zito Relova", "author_url": "", "post_date": "2018-10-11T03:38:23.080000", "content": "

Thanks for this. Really saved everyone a lot of time.

", "votes": 0, "replies": [] }, { "id": 402053, "author_name": "olivier", "author_url": "", "post_date": "2018-10-11T05:45:33.740000", "content": "

Thanks Jerome for your investigations ! This competition is a farce from start to finish....

", "votes": 11, "replies": [] }, { "id": 402067, "author_name": "rafamacalaba", "author_url": "", "post_date": "2018-10-11T06:12:58.413000", "content": "

Thanks Jerome for submitting!

", "votes": 1, "replies": [] }, { "id": 402069, "author_name": "mezoganet", "author_url": "", "post_date": "2018-10-11T06:19:21.527000", "content": "

@Jerome

\n\n

Thank you for submitting this \"joke\" !

", "votes": 1, "replies": [] }, { "id": 402071, "author_name": "dastoja", "author_url": "", "post_date": "2018-10-11T06:25:12.593000", "content": "

Thanks for sharing Jerome!

", "votes": 1, "replies": [] }, { "id": 402182, "author_name": "Sergey Bryansky", "author_url": "", "post_date": "2018-10-11T09:45:47.630000", "content": "

> Jerome Vallet wrote\n> \n> > I didn't submit it. I do not want to be accused of condemning the competition while many have worked a lot on it :))\n>

\n\n

Anyway, @emorej you did it;) The temptation is great in such cases! But thank you for sharing your insight and your honesty before community. I still have not started the competition as if I felt something after Santander did. So, now rule of the last kaggle is \"Before you start the competition data must brew\"!

", "votes": 0, "replies": [] }, { "id": 402327, "author_name": "olivier", "author_url": "", "post_date": "2018-10-11T14:31:20.787000", "content": "

Sergey, Jérôme did not do it by greed but simply because lots of kagglers (including me) asked him to check if his local CV would translate to LB.

\n\n

This also helps trying to get Kaggle to react on that since the rules don't excluded using Demo Account data...

", "votes": 3, "replies": [] }, { "id": 402351, "author_name": "Sergey Bryansky", "author_url": "", "post_date": "2018-10-11T14:54:53.787000", "content": "

Olivier, Maybe I missed something in this thread. So, my apologize. The most important thing is that the leak was opened early and there is time to solve this problem. But it remains an open question for me: \"Why Jerome and not somebody from Kaggle Team?\"

", "votes": 2, "replies": [] }, { "id": 402354, "author_name": "olivier", "author_url": "", "post_date": "2018-10-11T14:59:01.530000", "content": "

Good question Sergey ;-)

", "votes": 1, "replies": [] }, { "id": 402382, "author_name": "dylonLL", "author_url": "", "post_date": "2018-10-11T16:00:58.240000", "content": "

thanks jerome, you saved a lot of time for all of us. I am a beginner and this is my first competition so it is really sad that it turned out this way.

\n\n

olivier thanks as well, i learned a lot from your kernels.

\n\n

but out of curiosity i matched the visitid with the analytics client id and that alone increased the cv and lb score by a lot.

", "votes": 3, "replies": [] }, { "id": 402426, "author_name": "yilisg", "author_url": "", "post_date": "2018-10-11T17:43:40.207000", "content": "

Thanks Jerome -- at this rate perhaps I shouldn't even bother starting any competition until at least a month into it :-(

", "votes": 3, "replies": [] }, { "id": 403089, "author_name": "🐳", "author_url": "", "post_date": "2018-10-12T20:56:57.750000", "content": "

Same! I was pumped to get my hands dirty on this one.

", "votes": 1, "replies": [] }, { "id": 404469, "author_name": "Jeff", "author_url": "", "post_date": "2018-10-15T18:45:54.137000", "content": "

Thanks Jerome for exposing the leak. You saved a lot of time

", "votes": 0, "replies": [] }, { "id": 408635, "author_name": "DivyaanandSinha", "author_url": "", "post_date": "2018-10-23T07:41:44.960000", "content": "

I am new to this.. but can anyone tell me how is the revenue in the leaked data and totals.transactionRevenue related?

", "votes": 0, "replies": [] }, { "id": 417602, "author_name": "Ramona Fli", "author_url": "", "post_date": "2018-11-08T13:56:39.150000", "content": "

what score did you get without the leak?

", "votes": 0, "replies": [] } ] }, { "id": 401799, "author_name": "MichaelP", "author_url": "", "post_date": "2018-10-10T18:07:03.143000", "content": "

Maybe the point of this competition is that we all become more familiar with Google Analytics...

\n\n

@ogrellier - I have an analytics account I haven't looked at for a while. I was able to sign in to that and then access the demo account by by pressing the \"Access Demo Account\" button in the link Scirpus shared.

", "votes": 13, "replies": [] }, { "id": 401915, "author_name": "S D", "author_url": "", "post_date": "2018-10-10T21:48:07.250000", "content": "

I'll share my findings about the demo data... Jerome can confirm if I have the right insights.

\n\n

Go to User Explorer, enter the dates for the test data.\nClient Ids look like 1092690470.1510359633\nThe second part of the ID (e.g. 1510359633) corresponds to the time stamp (or visitID) of the first session of the visitor. That's how I was planning to link it to the test+train data. Sometimes you might not have the first visit in the data.

\n\n

If you click on a clientID, you will see the details of the sessions. There's a bigquery ID in there, but it doesn't seem to match with our fullVisitorID.

\n\n

Seem like you can download the rows 5000 at a time. I'm sure Jerome found a better way. Possibly you can filter by non bounce sessions.

\n\n

Initially I tried to only get the data with >0 sales. However I ran into visitors with missing sessions in the test data. I think that our test data is a weird subset of the original one. I even saw a visitor for which all but one of the sessions were removed.

\n\n

So, you might have more success instead if you focus on visitors with 0 sales. If sessions are removed from those, you still get sales=0...

", "votes": 10, "replies": [ { "id": 402142, "author_name": "Alexander Firsov", "author_url": "", "post_date": "2018-10-11T08:30:21.353000", "content": "

I did fast demo account overview, so from what I found, the procedure to get exact revenue should look like this:

\n\n
    \n
  1. On User explorer select test data timeframe, order by revenue, save user ids of paying users. Number of such users is less than 10_000, so you will need just two export each by 5_000.
    2. \n
  2. Click to each user to get User report and export the data, data has all details activities of the user.
    3. \n
  3. To map competition users and ga demo data user, use pageviews for the session, they are one to one, competition hits are sum of events and ecommerce events, and session start time which as I see is the same at least to minutes level.
    4. \n
  4. to get revenue find last purchase within a session (there can be several purchases), that is a 'totals.transactionRevenue' from competition data.
    5. \n
", "votes": 13, "replies": [] }, { "id": 403105, "author_name": "Antoine", "author_url": "", "post_date": "2018-10-12T21:57:02.840000", "content": "

For step 2, did you find a way to do that automatically for all 10 000 users ? Thanks for the tuto !

", "votes": 0, "replies": [] }, { "id": 403356, "author_name": "Alexander Firsov", "author_url": "", "post_date": "2018-10-13T12:58:32.227000", "content": "

Antoine, I did not try to automate this process. It is logical to use Analytics Reporting API instead of Google Analytics UI. Unfortunately Reporting API is disabled for demo account.

", "votes": 0, "replies": [] } ] }, { "id": 401955, "author_name": "eagle4", "author_url": "", "post_date": "2018-10-10T23:42:12.773000", "content": "

this competition is getting better (or worse) every day between the CV and now the leak... Another Santander (most recent one) competition style ?

", "votes": 3, "replies": [ { "id": 401957, "author_name": "S D", "author_url": "", "post_date": "2018-10-10T23:51:23.450000", "content": "

There might be a couple data points left once this leak is fully exploited...

", "votes": 0, "replies": [] } ] }, { "id": 401781, "author_name": "olivier", "author_url": "", "post_date": "2018-10-10T17:44:26.597000", "content": "

don't know, I already have an analytics account and it seems I can't open the demo account due to that !\nWe'll watch your progress on the LB ;-)

", "votes": 4, "replies": [ { "id": 401790, "author_name": "Scirpus", "author_url": "", "post_date": "2018-10-10T17:55:43.677000", "content": "

There does seem to be enough LB movement that I thought you released another kernel ;)

", "votes": 7, "replies": [] }, { "id": 401829, "author_name": "mezoganet", "author_url": "", "post_date": "2018-10-10T18:51:55.090000", "content": "

You are right, we are all waiting for another amazing Kernel from Olivier.

\n\n

I suspect some lazyness due to his comfortable rank ;-)

", "votes": 2, "replies": [] }, { "id": 401871, "author_name": "Dimosthenis Karaflos", "author_url": "", "post_date": "2018-10-10T20:14:29.317000", "content": "

Indeed, the forward sessions variables seem to boost the LB. But what worries me is that in my sense these kind of features are not appropriate. In a time frame of t we do not have information of t+1. Hence, when predicting a transaction value of the latest test set rows, we do not know if a visitor has visits that are not present on the test set. Thats means, that this kind of model has incomplete data to predict from...\nBy using only lag features i managed to obtain a CV of 1.538 but the LB is 1.4242. Olivier's kernel with higher CV obtains a much better LB score.

\n\n

This competition is a real puzzle !!! :)

", "votes": 4, "replies": [] }, { "id": 402066, "author_name": "mezoganet", "author_url": "", "post_date": "2018-10-11T06:10:55.850000", "content": "

@Dimosthenis Karaflos

\n\n

\"In a time frame of t we do not have information of t+1\"... So right !

", "votes": 1, "replies": [] }, { "id": 402183, "author_name": "Yao Zhongwei", "author_url": "", "post_date": "2018-10-11T09:47:17.663000", "content": "

From my understanding, when predicting a transaction value of the latest test set rows, , the forward session variable is likely to be NaT since we don't know whether there will be another session in the future, which is fine as it doesn't provide useful information. When it comes to other test set rows, we know the chance of purchasing is relatively small if another session of the visitor can be observed later. That information is quite valuable and should help to make better prediction. So from my perspective these feature engineering seems okay.

", "votes": 0, "replies": [] } ] }, { "id": 401936, "author_name": "holygo_1208", "author_url": "", "post_date": "2018-10-10T22:19:12.887000", "content": "

https://datastudio.google.com/u/0/reporting/0B2-rNcnRS4x5UG50LTBMT0E4aXM/page/nQN

\n\n

I discovered this when I tried to confirm the contents of the demo account. In Google DataStudio we can see the Google Merchandise Store GA report. I think that we can also label this with a correct answer. I may be wrong because I have not looked into this in detail. I'm seeking data analysis competition...

", "votes": 2, "replies": [ { "id": 401937, "author_name": "holygo_1208", "author_url": "", "post_date": "2018-10-10T22:20:22.493000", "content": "

It seems that we can also download the report by specifying the period.

", "votes": 0, "replies": [] } ] }, { "id": 401872, "author_name": "Allen", "author_url": "", "post_date": "2018-10-10T20:17:50.993000", "content": "

I was at first. But then I spent a good day and a half trying to see if there was anything useful from demo Account, but I haven't been able to find anything.

\n\n

The data from the demo Account doesn't match well with the data in the competition. For instance, the total revenue in the demo account for the training period is about 4 times the amount of revenue in the competition.

\n\n

Edit: I guess it turns out that there is a way to use the demo data. See Jerome's post for more detail.

", "votes": 2, "replies": [] }, { "id": 402001, "author_name": "Chinta", "author_url": "", "post_date": "2018-10-11T03:00:28.077000", "content": "

May be the organizers should stop this competition as it stands now. i.e. prior to leak. So that efforts put in could be saved.

", "votes": -5, "replies": [ { "id": 402002, "author_name": "", "author_url": "", "post_date": "2018-10-11T03:03:29.393000", "content": "", "votes": 0, "replies": [] }, { "id": 402005, "author_name": "Adri", "author_url": "", "post_date": "2018-10-11T03:09:11.970000", "content": "

I don't think it would be a fair solution as well. Some people efforts may not be reflected in the current state of the LB-PB and there was still 1 month remaining. In addition, a lot of the current medals would be given to people just submitting recent high scoring kernels.

", "votes": 5, "replies": [] }, { "id": 402029, "author_name": "yulia", "author_url": "", "post_date": "2018-10-11T04:05:58.830000", "content": "

That wouldn't be fair - many people are testing locally, and budgeting their time around work/life schedules knowing that the competitions was supposed to be going on for another month.

", "votes": 3, "replies": [] }, { "id": 402033, "author_name": "Chinta", "author_url": "", "post_date": "2018-10-11T04:14:11.970000", "content": "

Okay agree, that was one of the option. Other options are to exploit leak for the remaining time, or kaggle does something to fix data.

", "votes": 1, "replies": [] } ] }, { "id": 402295, "author_name": "Kain", "author_url": "", "post_date": "2018-10-11T13:23:59.257000", "content": "

Kaggle's turning into a sieve with these leak incidents in the recent months ;)

", "votes": 0, "replies": [ { "id": 402330, "author_name": "Regan Lui ", "author_url": "", "post_date": "2018-10-11T14:33:36.110000", "content": "

How are those incidents dealt with in the past ?

", "votes": 2, "replies": [] }, { "id": 402347, "author_name": "rafamacalaba", "author_url": "", "post_date": "2018-10-11T14:51:27.080000", "content": "

@Regan most recent leakage was for Airbus but I believe Santander was a data property and not a leak :)

", "votes": 1, "replies": [] } ] }, { "id": 402292, "author_name": "DimitreOliveira", "author_url": "", "post_date": "2018-10-11T13:19:04.930000", "content": "

Oh my, this are sad news.

", "votes": 0, "replies": [] }, { "id": 409878, "author_name": "", "author_url": "", "post_date": "2018-10-25T02:02:21.953000", "content": "", "votes": 0, "replies": [] }, { "id": 408072, "author_name": "", "author_url": "", "post_date": "2018-10-22T09:12:47.493000", "content": "", "votes": 0, "replies": [] }, { "id": 405110, "author_name": "", "author_url": "", "post_date": "2018-10-16T21:58:03.050000", "content": "", "votes": 0, "replies": [] }, { "id": 401911, "author_name": "", "author_url": "", "post_date": "2018-10-10T21:42:47.343000", "content": "", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "401880": "I stopped thinking about this competition for a few days when I realized that most of the revenues was quite easily recoverable from the demo... :(\nMy session level cv is 0.55 after integrating an extraction from the demo. \nAs I have already mentioned here [https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/68164][1] this morning, the days of the competition are unfortunately numbered unless Kaggle finds a solution for this data leak.\n\n\n [1]: https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/68164", "401759": "Anyone else worried how we use the demo data a.ka https://analytics.google.com/analytics/web/demoAccount\n\nWe know things like revenue per month etc so we are asked to predict things that we can almost look up!\n\n", "401799": "Maybe the point of this competition is that we all become more familiar with Google Analytics...\n\n@ogrellier - I have an analytics account I haven't looked at for a while. I was able to sign in to that and then access the demo account by by pressing the \"Access Demo Account\" button in the link Scirpus shared. ", "401915": "I'll share my findings about the demo data... Jerome can confirm if I have the right insights.\n\nGo to User Explorer, enter the dates for the test data.\nClient Ids look like 1092690470.1510359633\nThe second part of the ID (e.g. 1510359633) corresponds to the time stamp (or visitID) of the first session of the visitor. That's how I was planning to link it to the test+train data. Sometimes you might not have the first visit in the data.\n\nIf you click on a clientID, you will see the details of the sessions. There's a bigquery ID in there, but it doesn't seem to match with our fullVisitorID.\n\nSeem like you can download the rows 5000 at a time. I'm sure Jerome found a better way. Possibly you can filter by non bounce sessions.\n\nInitially I tried to only get the data with >0 sales. However I ran into visitors with missing sessions in the test data. I think that our test data is a weird subset of the original one. I even saw a visitor for which all but one of the sessions were removed.\n\nSo, you might have more success instead if you focus on visitors with 0 sales. If sessions are removed from those, you still get sales=0...\n\n", "401955": "this competition is getting better (or worse) every day between the CV and now the leak... Another Santander (most recent one) competition style ?", "401781": "don't know, I already have an analytics account and it seems I can't open the demo account due to that !\nWe'll watch your progress on the LB ;-)", "401936": "https://datastudio.google.com/u/0/reporting/0B2-rNcnRS4x5UG50LTBMT0E4aXM/page/nQN\n\nI discovered this when I tried to confirm the contents of the demo account. In Google DataStudio we can see the Google Merchandise Store GA report. I think that we can also label this with a correct answer. I may be wrong because I have not looked into this in detail. I'm seeking data analysis competition...", "401872": "I was at first. But then I spent a good day and a half trying to see if there was anything useful from demo Account, but I haven't been able to find anything.\n\nThe data from the demo Account doesn't match well with the data in the competition. For instance, the total revenue in the demo account for the training period is about 4 times the amount of revenue in the competition.\n\nEdit: I guess it turns out that there is a way to use the demo data. See Jerome's post for more detail.\n", "402001": "May be the organizers should stop this competition as it stands now. i.e. prior to leak. So that efforts put in could be saved.", "402295": "Kaggle's turning into a sieve with these leak incidents in the recent months ;)", "402292": "Oh my, this are sad news.", "409878": "", "408072": "", "405110": "", "401911": "" } }