Poker-AI.org

Re: Continual Re-Solving starting from Flop (DeepStack)

2018-04-10T13:56:04+00:00

I there really are no theoretical problems. I could compute the ranges and counterfactual values for each flop, I don't have them yet. But after thinking about it a bit, I don't think it's the best way after all.

Statistics: Posted by HontoNiBaka — Tue Apr 10, 2018 1:56 pm

Re: Continual Re-Solving starting from Flop (DeepStack)

2018-04-07T10:57:16+00:00

HontoNiBaka wrote:

Can I just use the flop round with the same algorithm, using the ranges that I get from my preflop strategy and initialize the counterfactual values that way at the start of the flop?

I don't 100% understand the DeepStack paper or your question, but I think you require both the ranges from the pre-flop and also the counterfactual values at the depth you want to go to. If you have the latter then you are good to go. Do you have them?

Statistics: Posted by spears — Sat Apr 07, 2018 10:57 am

Continual Re-Solving starting from Flop (DeepStack)

2018-04-05T16:47:28+00:00

I am thinking about creating a HU bot based on DeepStack techniques, the approach that I have in mind is this:
Preflop I play according to a strategy, that I computed by using Pio and which was postprocessed a bit using expert knowledge.
At the start of the flop I would use depth limited continual re-solving, which DeepStack used. When we want to solve the game from the root using that algorithm, the counterfactual values of each hand have to be set, this is done by solving the preflop round using the normal tree, not the gadget game.
Can I just use the flop round with the same algorithm, using the ranges that I get from my preflop strategy and initialize the counterfactual values that way at the start of the flop?
I know, that when we combine subgame solutions, we don't get a real Nash equilibrium, but do you think it would still produce a decent strategy?

Statistics: Posted by HontoNiBaka — Thu Apr 05, 2018 4:47 pm