This is a Python implementation of the actor-critic model of a rat in a watermaze, as presented in A model of hippocampally dependent navigation, using the temporal difference learning rule (Foster et al., 2000). Both RMW and DMP experiments can be simulated, and several kind of plots can be produced. However, the coordinate system is not part of this implementation.

This work is part of a small project done for an introductory course to machine learning applied to neuroscience, which was given at the École normale supérieure of Paris in 2019.

The code targets Python 3.7, but should run on Python 3.5+. It has the following dependencies:

• numpy (for the computations)
• matplotlib (for the figures)
• tqdm (for the progress bars)

They can be installed using pip (e.g. pip install --user numpy matplotlib tqdm).

The entry point of the code is main.py. It can be ran as a script with execute permission on an Unix system (if need be, you can assign it to the file by running chmod +x main.py).

The script accepts a few arguments, which are optional.

If you don't provide any, the default behaviour is to run both experiments 10 times (each), and to generate and save two kinds of figures:

• one figure per trial of the last run of each experiment (i.e 36 figures per experiment);
• one figure of the performance of the rat (average path lengths over all runs) per experiment.

Default behaviour:

./main.py

Simulate both experiments 50 times:

./main.py -n 50

Simulate the DMP experiment 20 times and skip trial figures:

./main.py --rmw 0 --dmp 20 --no-trial-plot

The code is mostly written in object-oriented style, and split in a few files with dedicated responsabilities. Each module is shortly described by the following table.