Code accompanying the paper "Contextualized Hybrid Ensemble Q-learning: Learning Fast with Control Priors".
We use Conda to create a Python environment in which the code can be executed.
Run:
conda env create --file environment.yml
conda activate cheq-env
The training for CHEQ and baselines can be started by executing main.py in the code directory, e.g. for starting the training for CHEQ-UTD20:
python main.py -algo "cheq" -G 20
For information about the arguments you can pass to main.py you can simply run
python main.py --help
to display a help page explaining the usage.
The logging of the runs is done with Weights and Biases. It is necessary to create an account and perform an intial login as described on their website.