Code used for the OpenAI Retro Contest Read more about Team Bobcats' journey in the eleven part blog series:
-
Days 4 & 5: Getting TensorFlow & Docker to work on my MacBook
-
Days 16–18: Running the PPO2 baseline code, and failing at TensorFlow & Docker optimization.
The explanation of the final code ( and the submission for the contest) can be found in improved-jerk.md
A list of the different tools that I made in the process can be found here: https://gist.github.com/tristansokol/062b1d509e2e8e6e250a30ae09928a58
All of the code is pretty much exactly the same as the final state of my working repository with the redactions of failed Q-learning model weights, an image of the sonic level and the sonic roms (for copyright concerns). The top level folders each represent a different agent attempt with our final agent being jerk_agent_for_understanding/jerk_agent.py
.
Much of the code is adapted from openai/retro-baselines which is Copyrighted by OpenAI