Closed aipoker39 closed 10 months ago
because in this case, the best exploitibility you can get is negetive, for player1, but for all player, it sums to be positive for sure.
What's the definition of exploitability? My understanding is that it indicates the distance between the strategy and nash equilibrium. How can the distance be a negative number?
What's the definition of exploitability? My understanding is that it indicates the distance between the strategy and nash equilibrium. How can the distance be a negative number?
In the above log,
player 0 exploitability means how much you can get from p0 if your strategy is optimal player 1 exploitability means how much you can get from p1 if your strategy is optimal
So it can be negative, because even if your strategy is optimal, you can still have a negative ev against a single player.
Since you can never know what a nash equilibrium is when running cfr , you can't calculate what you describe "the distance between stratgy and nash".
Thank you for the prompt and clear explanation; it has greatly helped my understanding. As I delve deeper into this repository and the codebase, having comprehensive documentation on the code and its underlying algorithms would be immensely beneficial. Regardless, I truly appreciate your efforts and contributions. Thanks again!
Thanks. Closing since the issue is resolved. Reopen if necessary.
Thanks for the project, I ran the solver using the default configs in README, and it output the following:
As you can see, the exploitability of player 1 is negative. What does that mean? Thanks in advance.