This repository contains simulation and analysis scripts to reproduce two figures from the publication "A closed-loop toolchain for neural network simulations of learning autonomous agents".
The network model is implemented in PyNEST and can be found in actor_critic_network/network.py
.
The YAML file provided should be used to set up a dedicated Python environment using Miniconda. After installing Miniconda the environment can be created:
$ conda env create --file environment.yml
Additional dependencies must be installed manually:
- install MUSIC
- install NEST with MPI and MUSIC support; since the models used in the manuscript are not yet available in the NEST master branch, you should use this branch instead
- install MUSIC-Adapters
Make sure to set your
PATH
,PYTHONPATH
andLD_LIBRARY_PATH
variables correctly.
$ cd figure_3_mountain_car
$ gymz-controller gym MountainCar-v0.json &
$ mpirun -np 6 music nest_mc.music
You might need to pass the option --oversubscribe
to mpirun, depending on your MPI library version.
$ cd figure_4_frozen_lake
$ gymz-controller gym FrozenLake-v0.json &
$ mpirun -np 6 music nest_fl.music