|
Post by steadyeddie on Dec 18, 2016 6:30:11 GMT -8
Sigh. In the carbon world I don't like simultaneous games. It's too random. People post justify their reasoning in games like Diplomacy or Power the game. In the silicon world I don't like the fact that it's easily possible to lose to Random at Blocker for example.
So, what do I do? Well I don't use a massive class I wrote to do find Nash equilibria. It kinda worked for one or two examples, but totally failed in many others, and I don't like my player to have lots special case code.
Instead, for simultaneous nodes (those with more than one player choosing), I keep stats for each player's personal moves, and then the code is very similar to a regular move. Except now I'm relying on the fact that for Nash-type problems, the inherant instability in the model will mean one time the players will pick one move, sometimes another. The model will then reinforce that behaviour. I'm sure they don't pick with the probabilities they are supposed to, but they do vary play randomly.
And that abomination is enough to be reasonably effective at games that should be factored, like C4 simul. The loss in performance is important, but SE can still join in.
|
|