An example to demonstrate the speed & simplicity of RlLib. Each example trains a
model, outputting results to .results
and videos of the last iteration to
.videos
.
This repository includes a devcontianer for one-click
setup. You can also install manually (pip install .
).
Cartpole is included as a comparison against multi-agent examples.
python ./src/singleAgent.py
This shows the performance of RlLib with tuned hyperparameters, solving Cartpole. Key stats are as follows (for hitting a reward of 990):
- Iterations: 42 (4000 steps per iteration, so c. 50 episodes by the end)
- Total time: 50 seconds (6-core CPU with GPU disabled)
python ./src/multiAgent.py
This takes around 100 iterations to get to a positive score (10 minutes on a 6-core CPU machine with 1 GPU).