README.md

Reward Modulated Self-Organizing Recurrent Neural Networks

PyPi package of RM-SORN: a reward-modulated self-organizing recurrent neural network: RMSORN-Paper

To install the latest release:

pip install rmsorn

The library is still in alpha stage, so you may also want to install the latest version from the development branch:

pip install git+https://github.com/Saran-nns/rmsorn

Usage:

Update Network configurations

Navigate to home/conda/envs/ENVNAME/Lib/site-packages/rmsorn

or if you are unsure about the directory of rmsorn

Run

import rmsorn

rmsorn.__file__

to find the location of the rmsorn package

Then, update/edit the configuration.ini

from rmsorn.tasks import PatternRecognition

inputs, targets = PatternRecognitionTask.generate_sequence()
train_plast_inp_mat,X_all_inp,Y_all_inp,R_all, frac_pos_active_conn = SimulateRMSorn(phase = 'Plasticity', 
                                                                                      matrices = None,
                                                                                      inputs = np.asarray(inputs),sequence_length = 4, targets = targets,
                                                                                      reward_window_sizes = [1,5,10,20],
                                                                                      epochs = 1).train_rmsorn()

Notebook is avaialble at RMSORN-Notebook