Quote:
Originally Posted by tombos21
I recently took a deep dive into the topic of Bayesian-adjusted HUD stats.
Method:
In short, I take the population data as a prior. Then adjust that prior using the observed sample. That gives us the posterior.
I create priors by splitting the population data into winning/losing players. Then I calculate the mean and SD of the population's HUD stats. Then I use a beta distribution to update the population stats with the player's observed data.
If we sample from the winning players, after 35 hands their 11/11 stats become:
- VPIP 22.8% ± 4.1%
- PFR - 18.7% ± 2.6%
My dataset comes from a nittier site. Looking at TBJ's data, it seems ignition is a lot looser. So my method has probably underestimated their stats!
Could you walk us through how you arrived at that conclusion, DDP?
I am definitely deferring to you here but I used Bayes Theorem so P(A B) = P(B A)*P(A)/P(B)
I wrote it all down here on page 3.
https://forumserver.twoplustwo.com/1...44/index3.html
I used your calculator to get the probability of being a nit and then probability of being reg and then plugged in the numbers from TBJ's data. I'm not sure I completely did it correctly so let me know if any of my inputs are incorrect.
Can you explain what the charts mean in laymen's terms? I appreciate you taking the time to do this.
So if he is a regular aka winning player. Then his VPIP is between 18.7% and 26.9% according to your data right? And if he is again a regular. His PFR is between 16.1% and 21.3%.
Is that right?
Would it be fair to say that the odds of him having a true VPIP of <20 is less than the odds of him having a true VPIP of 20 or higher? I'm trying to figure the odds of this guy being a nit (<20VPIP) or not. I'm claiming that the odds of him being an average regular is higher than the odds of him being a nit.
Also TBJ's data is from 50nl-200nl so on average there are less nits/more aggro players at higher stakes which also makes me think villain is more likely to be a regular than a nit.
Last edited by DooDooPoker; 05-04-2024 at 05:24 PM.