CFR Simultaneous

Attempt to work through An Introduction to Counterfactual Regret Minimization of Todd W. Neller and Marc Lanctot.

Compilation

make build

Usage

Examples

To run examples of CRF using default values

Rock-Paper-Scissor

To compute the average mixed strategies for both player in a RPS game:

make run_rps

Colonel Blotto

To compute the average mixed strategies for both player in a Colonel Blotto game:

make run_blt

More general case

The executable is located in .\build\bin\main.

Usage:
    main RPS [seed N]           # Run Rock-Paper-Scissor
    main Blotto [seed N B S]    # Run Colonel Blotto

Options:
    seed                        Seed used
    N                           Number of Monte-Carlo Steps
    B                           Number of battlefields (Blotto only)
    S                           Number of soldiers (Blotto only)

Visualization

A python script is included to visualize the convergence of the Monte-Carlo algorithm towards an average mixed strategy. It will create a gif using the iterated average strategies of player 1 during the Monte-Carlo algorithm. The exported data are located in export_avg_strategy.csv and the produced gif in average_strategies.gif.

Installation

make install

Execution

make anim

Example

For a Colonel Blotto with 3 battlefields and 5 soldiers using a seed of 20, we have the following average strategies:

Note concerning strategy indeces

Internally a strategy is encoded using a combinatorial number system. For instance the strategy 0 for a Blotto with 3 battlefields and 5 soldiers represents:

• 5 soldiers in the battlefield 0
• 0 soldiers in the battlefield 1
• 0 soldiers in the battlefield 2