Simple implementation of Echo State Network with NARMA benchmark task and RL with esn

basic echo state network implementation

problem setting

reimplementation of simple echo state network on NARMA-10 task. problem setting is from this paper

hyper parameters

All the hyper parameters can be controlled by config/baseline.yaml. If you want to try with other sets of parameters, all you need to do is modifying that yaml file.

As hydra, python library for hyper parameter control, is used, all the output of the program is going to baseline/ directory, specified with hydra:run:dir of yaml file.

code explanation

• esn_forward.py: calculate the states of echo state network
• narma_data_gen.py: code to generate narma dataset
• esn_offline.py: training and inference code for offline dataset
• main_offline.py: main code integrating everything above

how to run

necessary libraries

see requirements.txt

environment variables

add /path/to/esn_basic to $PYTHONPATH