re: PT stats, how many hands needed to see patterns?

[MicroBob]

i'm not very good at math at all. But I guess I understand the general concept of variance okay.

You are trying to find patterns in there.
Sometimes you are going to have a stretch where a certain hand holds up like that.

I bet I could go find any random sample of 5k hands and find similar abnormalities.

[uDevil]

[ QUOTE ]
I've had PT since last year, and was curious to know how many hands I need in my database before I can start seeing how my stats compare with a perfectly random distribution? Specifically, I have 5000 hands in my database on FTP and while my pocket eights have won every single time of the 12 times I've had them, my kings have won 65% of the time. That seems odd.

[/ QUOTE ]

Although I'm sure people mean well, there's very little useful information in this thread. The number of hands required to have a given level of confidence depends on the variance associated with the statistic you are interested in. A number like 100,000 means nothing without knowing what the confidence level is.

You may want to do some reading about confidence intervals.

You can find web apps that will do calculations for you. Here is one.

[creamfillin]

uDevil, I've always wondered about these concepts but never bothered to research them. I guess we're all a bunch of little kids blurting out big numbers.

[LionelHutz00]

some stats have higher standard deviations than others. vpip has a relatively low standard deviation, so your stats for vpip will converge to their true values relatively quickly. winrate has a high standard deviation so it takes many many many hands for that stat to converge to its true value.

[Sciolist]

Does anyone know how to calculate the standard deviation of PT stats? It calculates your win rate SD for you, so it's way easier to know your confidence interval. However, I don't know of a way to calculate the SD of VPIP, for example. I'm not maths expert, and I'm even a bit shaky on how to calculate SD in the first place, let alone creating a query to do it from the PT SQL DB.

I made a post about this on the probability forum a while back and got absolutely no replies.

[uDevil]

[ QUOTE ]
Does anyone know how to calculate the standard deviation of PT stats? It calculates your win rate SD for you, so it's way easier to know your confidence interval. However, I don't know of a way to calculate the SD of VPIP, for example. I'm not maths expert, and I'm even a bit shaky on how to calculate SD in the first place, let alone creating a query to do it from the PT SQL DB.

I made a post about this on the probability forum a while back and got absolutely no replies.

[/ QUOTE ]

You do need to write an SQL query. Which means you have to understand how the PT data tables are set up. I learned this myself by poking around a little using Microsoft Access. I queried columns for my player ID and the vol_put_money_in_pot column then took 10,000 records and copied them to Excel, where I just used the stdev and average functions to calculate the standard deviation and mean for my VP$P.

I got a mean VP$P of 25.2 (this is for SH LHE games) and a standard deviation of 43.4/hand. So after 1000 hands, I could have 95% confidence that my VP$P would be between 25.2+(1.96)43.4/sqrt(1000) and 25.2-(1.96)43.4/sqrt(1000), i.e. between 23.2 and 27.7.

For the OP, you'll be dealt a specific pocket pair one time in 221. The standard deviation (see binomial distribution ) for the number of times you'll be dealt a specific pocket pair (per hand) should be sqrt((1/221)*(1-1/221))=.0671. In 5000 hands, you can have 95% confidence that you will be dealt between 5000(1/221-1.96*(.0671)/sqrt(5000)) and 5000(1/221+1.96*(.0671)/sqrt(5000)) pairs of 8's, for instance. That is, between 13.3 and 31.9 pairs of 8's.

I'm not a mathematician and I make a lot of mistakes, so if these numbers are important to you, you should check them.

[uDevil]

[ QUOTE ]
You do need to write an SQL query.

[/ QUOTE ]

Actually, maybe you don't. If you have a sufficiently large number of hands, you can use your VP$IP% to estimate the SD of this statistic if you assume a binomial distribution. Then SD= sqrt(VP$IP%*(1-VP$IP%)).

So for me, SD=sqrt(.25*.75)=43.3, pretty close to the value I got before. That's pretty cool. [img]/images/graemlins/smile.gif[/img]

[Sciolist]

[ QUOTE ]
[ QUOTE ]
You do need to write an SQL query.

[/ QUOTE ]

Actually, maybe you don't. If you have a sufficiently large number of hands, you can use your VP$IP% to estimate the SD of this statistic if you assume a binomial distribution. Then SD= sqrt(VP$IP%*(1-VP$IP%)).

So for me, SD=sqrt(.25*.75)=43.3, pretty close to the value I got before. That's pretty cool. [img]/images/graemlins/smile.gif[/img]

[/ QUOTE ]
Great, thanks. I will have a go at this next week, let's see if I can get standard deviations & confidence intervals for all the stats...

[PBJaxx]

Check out the SSNL master sticky. Pokey wrote a great post on how to get good use out of poker tracker. There is also a post in SSNL discussing the affects of variance. I have not dug into this too much, but I have read that even the winrate SD that PT gives you can not be used in the ways we expect to be able to. The SD is actually based on wintrate/session (in terms of PT/100) as opposed to winrate/100 hands. If you dig around for this stuff there is more information on it including arguments as to which statistical models will work well for simulating the affects of variance on winrate.

I read enough to realize that there is a lot of work to be done there to get anything accurate or of true value. Instead, I use PT to review individual hands, and a few stats that converge quickly (such as VPIP). Good luck with this stuff, and I hope you pursue it further. I would suggest, though, that you use the search function, check the stickies, and browse the PT forums before going too much off on your own. A lot of very intelligent people have already put some work into this.

[El_Hombre_Grande]

It depends alot of what you are trying to measure, and how accurately. 8s aren't a better hand than KK, despite the 12 wins in a row that you have.

VPIP/PFR% and folding blinds, attempts to steal take less time to be meaningful than winrates or particular hands.

Add to the mix that your game and the games you are playing in are probably changing (anyone who doesn't evolve after 250,000 hands plain sucks) and you will come to see PT stats as measure of a process and a rough estimate of how you are playing.

Eventually, your positional stats and win rates from different positions will start to look reasonable (ie 55 is profitable in cutoff and button but not so in hijack) and then you will begin to be able to ask yourself real questions about whether you are overplaying/underplaying paricular hands from particular positions.

5 K is meaningless for that kind of analysis. Think 100,000 plus.