This is a mini-project related to the blog post.
It contains both our HRL interface and an example on how to use it.
The easiest way to use it in your project is to link directly to this repository
or using prebuild package (see Building a package
section below).
Install poetry.
Run:
poetry install
or
poetry install -E torch
if you'd like to run the example.
Use:
poetry build
The built distributions will be stored in the dist
directory.
To run tests use:
cd examples/maze
PYTHONPATH=. pytest
To train an agent use:
cd examples/maze
PYTHONPATH=. python train.py
By default logging to Weights&Biases is disabled. To enable it,
set log_to_wandb
to True
and update your credentials.
Below you can see training performance of the strategy agent.
And for the procedure setup (note 3x fewer steps than in the previous experiment):