Skip to content

thobotics/RoMBRL

Repository files navigation

Bayes-Adaptive Deep Model-Based Policy Optimisation

Prerequisites

Before installing the RoMBRL package, make sure the following Python packages are installed:

Other dependencies can be installed with pip install -r requirements.txt.

Running Experiments

Experiments for a particular environment can be run using:

python ALGORITHM
    --params     PARAMS  (required) Path to parameters file.
    --outputdir  OUTPUT  (required) Output directory.
    --logdir     LOGDIR  (optional) Path to log file.
    --policy	 POLICY  (optional) Policy type [bnn, lstm], default: bnn.

where ALGORITHM can be either online_bnn_trpo.py or bnn_trpo.py.

The default parameters that can be used to reproduce our results are stored under directory params/default/.

Trial data, dynamic functional samples as well as policy network will be saved each iteration under OUTPUT directory.

About

Code for Bayes-Adaptive Deep Model-Based Policy Optimisation

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published