The Data Mining Forum
This forum is about data mining
, data science
and big data
: algorithms, source code, datasets, implementations, optimizations, etc. You are welcome to post call for papers, data mining job ads, link to source code of data mining algorithms or anything else related to data mining. The forum is hosted by P. Fournier-Viger
. No registration is required to use this forum!
Date: April 06, 2020 12:01AM
I thought I would share some interesting data/results.
Some German casinos put their nightly roulette results on line, spin by spin in time order. You can get years of results if you want.
Here is one casino:
This is table by table with full statistics.
Now if roulette were truly random, this would be a waste of time. But people have been playing biased roulette wheels for 150 years, some with spectacular results:
There have even been academic studies on how to optomise bets/risk to maximise return in a biased wheel:
But there are a few problems, the wheels are hard to find--only 1 in 8 or 9 or more may have some bias, and you need LOTS of past results, many thousands. And because most casinos dont publish their results, you would need a team to clock the wheel. Even then, the fluctuations are enormous and you can get wiped out even if you have a slight edge.
A bigger problem is that the wheels have been getting better and better, they are low profile, very shallow pockets to allow the ball to hop out easily....all the while making the wheels more random.
Casinos now keep track of all the numbers that come up on all their tables and run them thru their own computers using the Chi Square test and 3 standard Deviations from average to indicate if the wheel needs to be sent in for maintenance.
So is it possible to find a playable bias that the casino is not aware of? It would seem impossible in this day and age.
When I learned about sequence mining, this application automatically came to mind.
I downloaded 90 days of complete spins in sequence (time order) from a wheel at Wiesbaden Casino and formatted it into SPMF format. I split it into 60 days and 30 days, this was my training data and my test data.
I did this for 3 of their busiest wheels.Training data has roughly 25K spins per wheel.
I have been testing different algorithms and checking results for the last 2 weeks. I am hesitant to post this because the results are amazing, I have rechecked things twice. I have 2 out of three wheels giving between 10-20 percent return on investment. This is very large if you consider blackjack card counters used to only get 1% advantage.
Different algorithms give different results and I havent optomised whether it is better to play 1, 3 7, 11 or 15 numbers, although playing too many increases the losses dramatically.
Now the really amazing thing is that by themselves, the single numbers DO NOT stand out! In other words, the casino chi square test on single numbers cannot see any problem. But in sequence, the best numbers have substantial profit.
I think this is definitely worth more study. There is a lot of work to do to study the fluctuations, optomise profit, how many numbers to play, best algorithm etc.
I don't think this has been done before.
I'll let you know if I have anything else to report.
Re: Roulette Data
Date: April 09, 2020 06:45AM
That is a very interesting topic. I have taken a bit of time to answer you because the last two days I have been very busy with teaching and other things.
I have read the story that you posted about the gambling professor. Very interesting story. He had a great idea at that time to study the wheels.
While I was working in Canada a few years ago, I talked with a student who did an internship at a company selling slot machines and other gaming machines. He told me that there was a statistician in their team to try to ensure that there was no bias in their machines and ways to take advantage of their machines. I do not know about how they carry their work. But your idea is very interesting. Indeed, if they dont check the sequential ordering, i imagine that their could some sequences of numbers that are more likely then others.
I am happy to know that you have been testing this with SPMF and found some promising results!
Looking forward to know more about it if you find something interesting.
I indeed assume that it would take quite a lot of work to find the best way of exploiting the patterns that you have found to build some good betting strategy around it.