Re:Lessons learned from a failed poker software business
by Retric on Tuesday November 18, 2008 @04:19PM
(#368738)
I suspect that the best long term stable strategie is to randomly assign a personal value for this hand. Something like .75 to 2x the actual value of your hand before the bet to reduce the amount of signal your sending out. If your both doing this and playing perfect poker then nobody wins.
Finding what the best random function might be would be an interesting test of machine learning. But in the real world few players are going to do something like this so finding out what their approach is is going to be a better approach.
If you make no estimate what kind of hand your opponent has you can only reason in three ways:
a) you assume he always has the best hand. So you throw away your hand.
b) you assume he always has the worst hand. So you put your chips in.
c) you bet big if you have good cards, and you throw your cards away when your cards are lousy.
Clearly (a) and (b) are bad strategies.
The last strategy is not as bad as the other two, but still terrible. More often than not your cards are bad. So your opponent will figure out you're not willing to make large bets with lousy cards, so every time you don't bet, he will, and every time you bet, he won't. You're going to lose your chips that way. Basic game theory.
So you need to figure out what cards he could have been dealt so that his actions make sense. Then, from that range of plausible cards, figure out how often he has a better hand than you do. If you like your odds, you put your chips in. Otherwise, you throw your cards away.
It's a lot more complex than this, of course. But the point is that "estimating your opponent's hands", as difficult as it may be, is absolutely crucial if you want to win.