Quote:
Originally Posted by bachfan
It's actually pretty simple conceptually; all that is required is to walk through the entire game tree, look at what cepheus does in every spot, consider the probabilities for each of his hands given our hands (is cepheus a guy? derail...), and choose the action that maximizes ev. With some clever engineering and a large cluster of computers, the maximum ev of a perfect opponent can be computed relatively quickly (at least for huhu lhe).
There is at least one paper out there that outlines some of the tricks to making this computation run fast... sorry I don't have it on me on the moment.
bachfan's got it.
The biggest trick is probably computing values for all hands in ~1000 operations instead of ~1,000,000 operations: if you did things in the obvious way, looking at the value for one hand requires looking at ~1000 opponent hands and there are ~1000 hands. Described in
http://poker.cs.ualberta.ca/publicat...t_response.pdf
pies01, if you were wondering about the specific advantage for the small blind and big blind, here's what we have. If you played perfectly as the small blind you would lose 87.718 mbig-blind/hand (4.3859 big blinds/100 assuming I can do basic math) to Cepheus. Playing perfectly as the big blind, you would win 89.691 mbig-blind/hand. Add that together, you would be ahead by 1.9727mbig-blind/2hands = 0.9863mbig-blind/hand.
(Just to make it clear, in this case by played perfectly I mean that you would pick actions which take the most possible money from Cepheus. You wouldn't have to play GTO to take Cepheus' money.)