Evaluating Meta-Reinforcement Learning through a HVAC Control Benchmark

This repositery contains code accompanying the paper- Evaluating Meta-Reinforcement Learning through a HVAC Control Benchmark. It includes all the code with comments for reproducing the experiments described in the paper, as well as for testing our new meta-RL algorithms.

The folder Meta-Exploration includes all our code for the variance seeking meta-exploration honours submission.

Dependencies

This code makes use of the open sourced stable-baselines implementation of ACKTR algorithm and can be installed in the way described below.

The code also requires TensorFlow v1.X

Installing stable-baselines

pip install stable_baselines