Poker-AI.org

Hi,

I am using public chance sampled CFR and just wanted to clarify how people calculate Regret at each Information Set.

Most of the research papers worked examples seem to use the terminal results for each information set (e.g. if the pot stands at 200 chips at showdown and I have bet 100, I am deemed to have won/lost 100 chips less rake multiplied by the associated probability of reaching that set etc for all sets).

I am wondering whether I should ignore bets made before reaching an information set, so in the above example if at information set 1 i bet 50 and then subsequently at 2 a further 50 and then we reach showdown, the utility component of information set 2 should be calculated as lose/win -50/150?

The thought process being that I can't influence the chips bet at previous information sets.

Any ideas would be appreciated!

Take a look at amax's code at http://poker-ai.org/archive/www.pokerai ... 335#p40335

thanks for replying - so in short example 1 is correct

It doesn't matter how you do it for CFR purposes (as for adjusting regrets you don't need to know if +200 from this point is -500 from the beginning as regrets are differences between best action at the moment and ev of other actions) but to make the code more clear and to allow for easier testing I like making all temp results absolute, so for example a fold on the turn after investing 300 on the flop is always -300 action and not 0 one.
You need to multiply by probability of reaching particular node though as noted in your OP.