All Categories

fix drop with boss clan

fix drop with boss clan

Search
Comments
Dec 24, 2021, 21:4212/24/21
06/20/19
936
Phantum

There was someone a few days ago just saying to send data from a single clan for 2 weeks and they could prove/disprove ...  Surely that is not randomly sampled?  

There are a large number of users reporting data for the full 2 months. Data is linked to an anonymous user profile so you can follow each user over 2 months. and see statistically if users stop reporting when they get a sacred... 

If the data was biased due to non-random sampling then I am sure you could show this in the first month of all users data.  The bias would show if you compared the histogram to the expected distribution from the data-mined model.   

Remember this initial experiment was to show users that their drop data did conform to the data mined model as in their first report.  The bimodal distribution came out as unexpected in the second month with many of the same user's data.  Not just an injection of new users data as I understand it.   So unless you think the users deliberately falsified data then how can some users continue to get normal results when a statistically significant change in the distribution for the next month.

Remember there is a mathematical model with probabilities coming from data mining which is generally accepted by the community (although no information was released enabling users to verify this).  This gives you the whole population statistics.

The vast majority of users report chest drops that agree with the predictions of the data-mined model for the first month.  Most users report the same for the second month.  A smaller but (significant depending on your test) report a changed drop rate distribution on the second month.

You can filter the raw data using your own skills and professional experience.  Only take UNM chests, reject users who give an impossible data value such as 6 saceds (max 2 from 2 chests). If you think users stopped giving data when they got a sacred drop then remove users who conform to this pattern.

Any professional or university student would have access to mathematical analysis software (eg MatLab stats package) which can very quickly perform various hypothesis tests.

1 That the results were all produced by the data mined mathematical mode for the 1st and 2nd months.

2. That some users drop probabilities changed in the second month to form the bimodal distribution that can be represented by the data-mind probabilities and a single second set of probabilities.

If someone could show a mistake in the data or the analysis then it would be great, problem solved.  Surely there is more that can be done than just dismissing it?

"There was someone a few days ago just saying to send data from a single clan for 2 weeks and they could prove/disprove ...  Surely that is not randomly sampled?  "

You did not hear me saying that!  100% agreed this would not be a random (or sufficient) data.

Dec 24, 2021, 22:0412/24/21
Dec 24, 2021, 22:04(edited)
12/19/19
3821
Angwil

"There was someone a few days ago just saying to send data from a single clan for 2 weeks and they could prove/disprove ...  Surely that is not randomly sampled?  "

You did not hear me saying that!  100% agreed this would not be a random (or sufficient) data.

That was me, lol, though I don't think that is exactly what I said. But I would argue that a full clan of any duration with complete data would be far superior to what we have so far. 

But hey, I have joined the Russian pov didn't you see :)

- Comrade trips


Dec 24, 2021, 22:1412/24/21
06/20/19
936
Trips

That was me, lol, though I don't think that is exactly what I said. But I would argue that a full clan of any duration with complete data would be far superior to what we have so far. 

But hey, I have joined the Russian pov didn't you see :)

- Comrade trips


In a sense you are correct.  You need reliable data from the population at large, then you need data from the population you are trying to demonstrate is significantly different from the population at large.  The assumptions would include that the particular clan population was somehow different from the population at large.  This subsumes that the 18% of affected players are seperated by clan, rather than geography, spending, etc.  All you could rule out is that that particular clan was not significantly different from the population at large.  But you'd have to have the data for the population at large as well.

Dec 24, 2021, 22:5312/24/21
12/19/19
3821
Angwil

In a sense you are correct.  You need reliable data from the population at large, then you need data from the population you are trying to demonstrate is significantly different from the population at large.  The assumptions would include that the particular clan population was somehow different from the population at large.  This subsumes that the 18% of affected players are seperated by clan, rather than geography, spending, etc.  All you could rule out is that that particular clan was not significantly different from the population at large.  But you'd have to have the data for the population at large as well.

Basically. Tbh, at least with uninterrupted data from one entire clan we can rule out some self selection bias and selective reporting. But this is still self reported survey data, always crap in my experience:)

Basic analysis of limited clean data > rigorous advanced statistics on crap

This is why statisticians often get in the way of business strategists :)

Dec 26, 2021, 08:3512/26/21
07/17/20
19

Hi Sorry, my posts are a general reply to all I have read and I don't always remember who said what.  I also usually only have a short time at the computer to write.

I am interested in others analyses because the only tools I have are google sheets and javascript since I am no longer with a university or government institute.  As I said my tests were limited to looking for an anomalous number of people getting zero Sacreds at all, as I have.  I only looked at people getting zero sacreds over n drops and found no such anomaly. 

In the bimodal distribution reported there was a reduced chance of getting a sacred in the group getting more ancient books.  However, using google sheets or any other online software on such a large data set is just too slow for me to continue.

I remain convinced that anyone with the appropriate software could test the data from the Bot for tampering because there is an analytical model that gives the accurate population distribution to test again.  I think that most of the limitations talked about are more for situations where you do not have the actual population distribution but are trying to infer it from a sample.

Hence my case against non-random sampling. How likely is it that some bias in the sampling was able to produce not only the bimodal distribution in the second month, but also from the expected population distribution in the first month?  It would have been different if the first months data was not included.

Dec 26, 2021, 08:4912/26/21
07/17/20
19

Oh and as for where I want to go with this...  I just saw some curious data and I would like an explanation.  For me the most interesting explanation would be to see someone prove evidence of sampling error or demonstrate some mathematics I could learn.  Or for someone to admit to some hoax I guess.  I very much doubt that Plarium would admit if there was a problem, given how far this has gone I would expect it would just vanish in some future patch...

Similarly for my own record of not getting any sacred in over 50 drops now...  I think I would actually prefer that Plarium proved to me that I was just unlucky, like 1 in 40k or whatever it was....  I have managed these fusions without any sacreds, partly by learning to hold on to claim any rewards from fusions and shards until they count towards a champion chase.  Rewarding me with 10 sacred shards I missed out on is would just be one click and likely not give me one of the top rank legendaries or Epics that would make a big difference.

The only reward that would make a real difference for me would be Plarium offering instead to do the grind in my account for 1 week to give me a holiday where I could spend my spare time instead trying to complete a new faction wars boss.

Dec 26, 2021, 12:4012/26/21
Dec 26, 2021, 15:18(edited)
06/21/20
7

Hi. We have some updates!

We suppose, that this case can be closed. Dec 24th we detected, that UNM CB drop rates got back to normal. So now we have drop as usual.

Truly speaking we need about 4 weeks to make sure, that actual chances (after rollback) are the same as before Nov 17th. But we believe that everything will be okay.

Plarium claimed that there were no changes, and in theory it can be true, but the chances are extremely low. We think that most relevant version is A/B test.

We wish you happy holidays!


i


i


harleQuinnModerator
Dec 26, 2021, 14:3412/26/21
02/24/19
2630

Thank you very much for the update. Happy Holidays to you as well. 

Dec 26, 2021, 15:2912/26/21
07/17/20
19

Thank you for the update.  I really appreciated your work and sharing your data with everyone.

It is really frustrating that we may never know for certain although I personally think something was up!  maybe one day an insider from one side of the argument or the other own up!

Was there a client update I missed at the time it changed back?  Or any way to tell about server updates from a version tag in the data mining?

harleQuinnModerator
Dec 26, 2021, 15:4412/26/21
02/24/19
2630
Phantum

Thank you for the update.  I really appreciated your work and sharing your data with everyone.

It is really frustrating that we may never know for certain although I personally think something was up!  maybe one day an insider from one side of the argument or the other own up!

Was there a client update I missed at the time it changed back?  Or any way to tell about server updates from a version tag in the data mining?

Gorl and Masha's posted data has no data mined information, just drops logged by users over time. Nor is such data mined stuff really allowed on the forums or the official discord, I have been told.

There was a server reset, and it was announced in the discord. :)

Dec 26, 2021, 15:5912/26/21
06/21/20
7

Usually you don't need server restart to switch off a/b exp.

Exp config looks like {"id":"exp-111", "data":{"rate":0.1, ...}} or something like this. Server asks exp-system: "which exp is enabled for user with id-1234?". And gets back exp-id. You need server update to be able to handle exp with defined id. (Without update game-server does not know that it should change drop prob if exp-111 is enabled)

To stop exp you should just remove (or set expire date) exp config from exp-system. Usually you can do it via web ui.

Ofcourse implementation can be different, but generally it works like I described.

*Also it means that start of exp can be much later then related server update.

Dec 26, 2021, 16:3512/26/21
12/09/21
22

All your numbers all your graphics all your bots

 ARE BASED ON CHANCE

there is no way to prove that drop was changed even if it was

Dec 26, 2021, 17:1012/26/21
06/21/20
7
12313

All your numbers all your graphics all your bots

 ARE BASED ON CHANCE

there is no way to prove that drop was changed even if it was

¯\_(ツ)_/¯

harleQuinnModerator
Dec 26, 2021, 17:2212/26/21
02/24/19
2630

After talking to Gorl and Masha, I'm closing down this thread since they have nothing new to report, and the discussion here is just going in circles.

Thanks everyone!

The topic is locked. You cannot post comments.