Using a genetic algorithm to evolve a good qwopper.

Grab the docker container from here via docker pull nathandemaria/qwop_master or build your own with docker build master/.

Run with docker run -d -v genealogy:/root/master/genealogy nathandemaria/qwop_master, or your own tag :)

From the directory containing docker-compose.xml, run docker-compose up -d --force-recreate. This will start two docker containers - a Selenium server running Firefox, and an R container running a listener that gets tasks from the QWOP master, plays a round of QWOP, grabs the score using Tesseract OCR, and sends it back. Look here for the definition of the R listener.

I used the amzn-ami-2016.03.e-amazon-ecs-optimized image, with a t2.micro instance. Run the master docker container with docker run -d -p 5000:5000 -v genealogy:/root/master/genealogy nathandemaria/qwop_master:20160730, and use the Public IP as QWOP_MASTER_ROOT in the slave step. Make sure to open up port 5000, at least to the slave auto scaling group created below.

Create an ECS cluster, service, and task for QWOP. Thanks to Micah Hausler's container-transform docker container for translating my docker-compose.yaml to AWS's JSON format. See ecs-task.json for an example config (don't forget to replace <QWOP_MASTER_ROOT>). I attached an auto scaling group using the ECS optimized AMI (amzn-ami-2016.03.e-amazon-ecs-optimized), on t2.small instances.