Evolution Strategies for Board Games

How to run the code?

For using fully connected layers
python experiment_runner_modified.py --mutation SM-G-SO --mutation_mag 0.01 --display --domain connect_four --max_gen 200 gen_length 100 --pop_size 150

For using convolutional layer
python experiment_runner_modified.py --mutation SM-G-SO --mutation_mag 0.001 --display --domain connect_four_CNN --max_gen 200 gen_length 100 --pop_size 150

For population play
python experiment_runner_modified.py --mutation SM-G-SO --mutation_mag 0.01 --hidden 256 --layer 1 --population_play --state_archive --max_gen 200 --gen_length 100 --pop_size 150 --fast

To find about all the arguments use:
python experiment_runner_modified.py --help

TODO:

• Resolve bug in CNN code.
• Add some descriptive figures.

I have also written a blog, which explains what we have done in this project. Here is the link to the blog. You are welcome to create issues for your queries regarding the project.

This code is based on Uber-AI safemutation work. You can find the original code here.

The second part of my internship was on Evolution Strategies for Multi-Armed Bandit problem. Please do checkout Evolution Strategies for Multi-Armed Bandit problem.