Reinforcement Learning Durham

A Python package for the Reinforcement Learning courses at Durham University.

See the notebooks and gallery for some examples.

Installation

Start with a clean Python 3.10 environment (recommended) and install/upgrade via terminal as

pip install swig
pip install --upgrade rldurham

or Jupyter notebook as

!pip install swig
!pip install --upgrade rldurham

Known Issues

• Have you tried turning it off and on again? Restarting your kernel and/or restarting the install from a fresh and clean Python 3.10 environment resolves most problems.
• Check the error messages! In many cases, they provide useful information and in many cases the problem is not with rldurham but either a general Python problem or a problem with gymnasium.
• On NCC you need to create your own custom environment/kernel (as for the deep learning coursework); you cannot pip install ... things in the default environment (it may first look as if you can, but then the packages cannot be found).
• Problems related to swig
  • Remember that swig has to be explicitly installed before rldurham (as above) because of dependency issues with the packages used by rldurham, in particular gymnasium.
  • Under python 3.11, swig may need to be installed using system tools (i.e. not via pip), better avoid 3.11 and use 3.10 instead.
  • If you see errors mentioning Box2D, this is related to gymnasium, which requires Box2D, which requires swig, which frequently causes problems. Installing from a clean environment, first installing swig, and using Python 3.10 (avoiding higher versions) are the best ways to avoid these problems.