Implementation of Complementary Temporal Difference Learning model

Using Stable-Baselines 3

Thanks to Sam Blakeman and Denis Mareschal

requirements.txt contains the modules required by the project

pip install -r requirements.txt

Latest versions of libraries should work, however there may be a need to downgrade numpy<2

pip install "numpy<2"

MainCode folder contains the Python modules used.

Primary Files

TrainCTDL.py - main code to run models and features many test functions used during development. Run with quicktestCTDL() function to run against Maze 1 and ensure all elements are installed correctly. Output will feature in a Results folder with a sub-folder incorporating date and time.

CTDL.py - main code for CTDL model, derived from Stable Baselines DQN class

CTDLPolicy.py - implements policy and predict function using both DQN and SOM contents

SOM.py - implementation of SOM and SOMLayer classes

Maze.py - implementation of Maze in Gymnasium environment

Plotters.py - code for outputting graphics

ExplanationUtils.py - extraction of explanations. Can also be run standalone to extract explanations from previous models.

EventsForModel.py - not used, but can be used to hook into CTDL events