search

Jamesflynn1 / CS344-Opponent-Exploitation-Poker

A third year uni project aiming to implement and evaluate the EFR algorithm with different deviation types and explore a potential tradeoff between exploitability and expected value of a strategy in practice.
0 stars 0 forks source link

Proper understanding of limitations of current methods #15

Closed Jamesflynn1 closed 1 year ago

Jamesflynn1 commented 1 year ago

Understand how current methods are limited and suggested improves that could be made to them

Jamesflynn1 commented 1 year ago

EFR and CFR still assumes perfect rationality, we know this is not the case so we can apply opponent modelling to understand how the other players will deviate from the computed equilibrium, this will be a step performed after the EFR compuations.

L2 minisiation performs well (better than L1 in most regards in terms of expected wins).

See #10

Jamesflynn1 commented 1 year ago

EFR hasn't had many performance enhancements applied, Monte Carlo methods and function approximation could allow speedups, even sufficiently to expand the set of behaviour deviations, note what will the effect of this be, why would we want to do this.

See #20

Jamesflynn1 commented 1 year ago

Abstraction is required for most forms of Poker to be tractable, are current methods sufficient, do they lose valuable information in the process, could more be abstracted away.

See #19

Jamesflynn1 commented 1 year ago

This issue was aimed to guide research and develop of new techniques for the project. Sufficient progress has been made on this topic for the progress report. More will be required for opponent modelling / exploitation.