This directory contains the configuration files used to reproduce the results in my thesis. See the rlglueexp module for information on how to run them.
The files contain experiment specifications for 4 different reinforcement learning agents each run in 3 different RL environments. Every agent and environment combination is run 10 times. Experiments are grouped by environment, namely:
- mean_acrobot.config for the acrobot environment;
- mean_car.config for the mountain-car environment;
- mean_pole.config fo the cart-pole environment.
The reward obtained during each of the experiment instances is logged in a file placed in the respective instances folder. To plot the reward curves obtained by each of the agents use the analyze.py script.