This project is part of a competition in the course "Decision making under uncertainty" homepage: http://www.cse.chalmers.se/~chrdimi/teaching/optimal_decisions/index.html
Team: Oscar Carlsson, John Karlsson, Oskar Lindgren
This project is explained in detail in the report found in report
Keywords: Discrete state space, Implemented agent using Sarsa-lambda, KL-UCB, option to vary parameters
This plot illustrates how our agent performs on average over 15 runs in a experiment with 1000 episodes. ![Plot] (https://raw.github.com/Oscarlsson/RL-competition/master/data/100episodes_50runs.png "Our agent against several different environments")
- RL-glue 3.04
see RL-GLUE/install.sh
and [rl-glue] (http://glue.rl-community.org/wiki/Main_Page "rl-glue") for more information
- C++11 (Some environment)
- matplotlib (Run.py and plotresults)
- pandas 0.12 (Run.py and plotresults)
The code containing the Agent Experiment Environments
is found insrc
run using cd src; ./run.py
to run the default setup defined in etc/runpyconfig
Run ./run.py -h
for a list of available arguments
output is stored in ../outputs
and you can generate pretty plots using
./plotresults.py -D ../outputs/<yourrun>