This repository contains the code for the GIL project on general gameplay through quailty-diversity search

A typical install and run might look like:

• git clone https://github.com/bwindsor22/quality-diversity-rl.git
• cd quality-diversity-rl
• bash install_bam4d.sh
• export PYTHONPATH=`pwd`:$PYTHONPATH
• python evolution/run_mapelite_train.py --num_threads 9 --num_iter 50000

• There are two gvgai libraries. install.sh installs the ruben library, which was tested mac. install_bam4d.sh installs the bam4d library, which was tested on unix.
• testing the bam4d gvgai library: Run python -m unittest test/test_bam4d.py. If you don't see commands and actions getting printed out, try these steps:
  • Check java: try conda install -c conda-forge/label/gcc7 openjdk to get java. Check java --version shows java. Ensure which java points to the same java as echo $JAVA_HOME. $JAVA_HOME/bin/java should be the same as which java
  • Check gradle: cd GVGAI_GYM_BAM/GVGAI_GYM/ and run ./gradlew --status
  • If you've changed gradle or java, try reinstalling bam4d
  • Bam4d has only been tested on linux. If on Mac, run the Ruben library with install.sh
• Common errors
  • module evolution not found: make sure you set your python path, above
  • module DQN has no attribute conv1: ignore, this sometimes happens when starting up a lot of threads at once, but future threads will succeed

• check ganglia (ask on slack for details) to watch your CPU usage. Ideally, you want to see system and user usage as high, but that wait is zero. If you see wait, you're overloading the cpus
• I recommend running from tmux and using one pane to run the script, while having htop open in another pane