Implementing the KernelUCBVI
algorithm from the paper. The algorithm and the baselines are implemented in the folder algorithms/
. The folder config/
contains the parameters defining the experiments.
-
Requirements:
- Python 3.7
rlberry
version 0.1- pyyaml
-
Create and activate conda virtual environment (optional)
$ conda create -n kernel_ucbvi_env python=3.7
$ conda activate kernel_ucbvi_env
- Install requirements
$ pip install 'rlberry[full]==0.1'
$ pip install pyyaml
- Run
Kernel-UCBVI
andAdaptiveQL
experiments:
$ python run.py config/experiments/twinrooms_exp.yaml --n_fit=8
- Run
UCBVI
andOptQL
experiments:
$ python run.py config/experiments/twinrooms_exp_unif_discr.yaml --n_fit=8
- Generate the plots:
$ python plot.py