Poker-AI.org
http://poker-ai.org/phpbb/

Help with amax RiverSolver
http://poker-ai.org/phpbb/viewtopic.php?f=24&t=2658		 Page 1 of 1
Author:  nemesis17 [ Fri Dec 06, 2013 12:46 pm ]
Post subject:  Help with amax RiverSolver
I was trying the amax River Solver (it's the RiverSolver2.zip taken from here http://www.poker-ai.org/archive/www.pok ... a&start=40) with the following board and game tree:



Code:
 

        var board = Hand.ParseHand("2d 6s 8c 7s Ah");

        static Decision CreateGame4(HoleDistribution[] d)
        {
            return
                new Decision
                (
                    0, d,
                    new Decision(1, d,
                        new Showdown(190),
                        new Decision(0, d,
                            new Fold(-190),
                            new Showdown(190 + 230))),
                    new Decision(1, d,
                        new Fold(+190),
                        new Showdown(190 + 230)));


        }
 


However I am getting very odd results. The following is line 3260 taken from the attached solution dump. It shows that player 0 is check/folding the nuts 28% which makes no sense.
Code:
Ts 9s 28 72 | 0.25 0.633300114817056 | -9768030032.41891 150030


Anyone knows what could be wrong here?

Thanks.

Attachments:
solution.zip [30.46 KiB]
Downloaded 481 times
Author:  nemesis17 [ Fri Dec 20, 2013 4:19 pm ]
Post subject:  Re: Help with amax RiverSolver
I figured that there is nothing wrong with the solution except that I didn't interpret the strategy output correctly.
This particular hand is never checked in the first decision node so it never reaches the check-fold leaf obviously.
However I don't understand why the normalized average strategy is not zero in the check-fold and check-call leafs in this example.
Could anyone please explain?

Thanks.
Author:  proud2bBot [ Fri Dec 20, 2013 7:47 pm ]
Post subject:  Re: Help with amax RiverSolver
In the first iteration, every path of the gametree is reached with every bucket because you have equal probabilities for each possible action...
Author:  nemesis17 [ Fri Dec 20, 2013 8:53 pm ]
Post subject:  Re: Help with amax RiverSolver
proud2bBot wrote:
In the first iteration, every path of the gametree is reached with every bucket because you have equal probabilities for each possible action...


I see and given that the algorithm keeps adding the cumulative strategies the average strategy will never be zero even for unreachable states right?
Author:  proud2bBot [ Sat Dec 21, 2013 2:02 am ]
Post subject:  Re: Help with amax RiverSolver
correct. you can easily check it: in your case, the cumulative strategy values in this node should be very low (i.e., only visited at the beginning of the learning until the algorithm realized it should never check).
Page 1 of 1 All times are UTC
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/