Getting started

Install

pip install echostatenetwork          # core (numpy/scipy/matplotlib)
pip install "echostatenetwork[opt]"   # + scikit-optimize for Bayesian hyperparameter search
pip install -e ".[dev]"               # development

Minimal workflow

import numpy as np
from echostatenetwork import EchoStateNetwork

y = my_time_series                       # (Nt, N_dim)

esn = EchoStateNetwork(y=y[:1].T,        # sets the state dimension
                       dt=0.01,          # time step of the data
                       N_units=200,      # reservoir size
                       t_train=40.0,     # training window
                       t_val=4.0)        # validation window
esn.train(y)                             # ridge regression + Bayesian hyperparameter search

Training accepts a single trajectory (Nt, N_dim), a batch of segments (L, Nt, N_dim), or a ragged list of segments of different lengths. After training, esn.step(u, r) advances the reservoir one step; washout (open loop on observed data) followed by closed-loop prediction is shown end to end in the tutorials.

Parametric ESN

Pass input_parameters (shape (N_param, L), one parameter vector per training segment) to condition the reservoir on a physical parameter — see the parametric ESN tutorial.

Parallel ESN

For a spatially extended system, such as Lorenz-96 on a ring of sites, pass patch_size (the sites of a patch), halo (the sites read on either side of a patch) and periodic to run one reservoir of N_units units per patch. By default, the patches share the input, reservoir and readout matrices and the hyperparameters; shared=False gives every patch its own network (patches). The reservoir state stacks the patches (N_r = N_patches * N_units rows), and state_positions gives the position of every state row for covariance localization — see Parallel echo state network.

esn = EchoStateNetwork(y=y[:1].T, dt=0.02, N_units=300,
                       patch_size=2, halo=3, periodic=True)       # shared=False: independent patches