API reference

Layers

Layers available for inputs, reservoirs, readouts, etc.

The objects must be explicitly imported:

import water_tank as wt

rc = wt.layers.RecurrentLayer(size=200, tau=3.0)

RecurrentLayer	Reservoir of recurrently connected neurons.
LinearReadout	Linear readout layer. Performs a weighted sum of its inputs, without dynamics.
StaticInput	Static placeholder for input vectors.
TimeSeriesInput	Dynamic placeholder for series of input vectors.

Connecting layers with each other.

Connects two layers with a (sparse) weight matrix and optionally a bias vector.

Learning rules for online training of a projection.

The objects must be explicitly imported:

import water_tank as wt

lr = wt.rules.RLS(projection=esn_rc, delta=1e-6)

DeltaLearningRule	Delta learning rule (online linear regression).
RLS	Recursive least-squares (RLS) learning rule for FORCE learning.

Simple wrappers around numpy’s random distributions.

The objects must be explicitly imported:

import water_tank as wt

values = wt.random.Uniform.uniform(-1., 1.).sample((10, 10))

Const	Constant “random” distribution, returning the same value.
Uniform	Uniform distribution, returning values between `min` and `max`.
Normal	Normal distribution, returning values with a mean of `mean` and a standard deviation of `std`.
Bernouilli	Bernouilli (binomial) distribution, returning the first of the two values with probability p.

Various tools to facilitate simulations.

Data structure to record activities during a simulation.