|
#1
|
|||
|
|||
Re: Computer Programmers solve checkers.
The reason it sucked is precisely because it did the +EV play (in terms of its hand's pure value) every time. This is an exploitable strategy. If the computer never bluffs, it's not going to get maximum value for its big hands against a decent player. Instead, it should take into account how often its opponent will fold to a bet/bluff and how much is in the pot, and then throw in an appropriately-sized bluff bet randomly a certain percentage of the time to mix things up. According to game theory, there's an optimum bluffing frequency (and bet size) for each opponent, which makes it impossible for your opponent to beat you in the long run. (See Sklansky's Theory of Poker for why this works.)
|
#2
|
|||
|
|||
Re: Computer Programmers solve checkers.
[ QUOTE ]
[ QUOTE ] I made a computer opponent that did the EV+ play every time and it sucked. Easily exploitable. [/ QUOTE ]The reason it sucked is precisely because it did the +EV play (in terms of its hand's pure value) every time. [/ QUOTE ] This is better known as not making the +EV play, e.g., failing to recognize and make profitable bluffs. This is the 2+2 theory forum. Elsewhere, let the ignorant masses say poker is nothing but psychology, but at least here, we should be able to accept the idea that bluffing might be mathematical, as was analyzed in papers by Borel and von Neumann in the earliest modern work on game theory. Don't blame the failures of your program on the inadequacy of +EV poker. |
#3
|
|||
|
|||
Re: Computer Programmers solve checkers.
My bot was primitive and simple, but based on exhaustive simulations at each street, and it sucked pretty hard, but this is due to it's simpleton nature betting and calling when it was >50% to win the hand and not taking ranges or history into account.
|
Thread Tools | |
Display Modes | |
|
|