ZeroSumGameSolve

This repo contains code for our submission titled Second-Order Algorithms for Finding Local Nash Equilibria in Zero-Sum Games.

We use Julia for our implementation. To run this code,

1. Open the Julia REPL.
2. In the REPL, type ] to open package mode. Then enter the command activate . .
3. Exit package mode by hitting backspace.
4. Run import Pkg, Pkg.instantiate() to download all dependencies. You may need to run Pkg.resolve() before.
5. You are now ready to run experiments.
6. To generate unconstrained toy example results: run include("experiments\\ToyExample\\src\\RandomToyExample.jl").
7. To generate constrained toy example results: run include("experiments\\ToyExample\\src\\TestConstrained1.jl") and include("experiments\\ToyExample\\src\\TestConstrained2.jl"). To generate additional result given in appendix: run include("experiments\\ToyExample\\src\\ToyExample.jl"). To generate plot of CESP diverging: run include("experiments\\ToyExample\\src\\TestCESP.jl").
8. For the GAN example:

• Go to the GANExample directory in the terminal: cd experiments/GANExample/ and start Julia REPL with julia.

• Activate the environment via ] and activate ..

• Precompile the package: using GANExample.

• To run training for GDA: GANExample.train_gan_standard().

• To run training for LSS: GANExample.train_gan_zero_sum(; approach = "mazumdar").

• To run training for our approach: GANExample.train_gan_zero_sum(; approach = "ours_optimizer").

• To reproduce the result plots: GANExample.plot_gan_example_comparison().