reservoirpy.compat.regression_models.RidgeRegression#

class reservoirpy.compat.regression_models.RidgeRegression(ridge=0.0, workers=-1, dtype=<class 'numpy.float64'>)[source]#

Ridge regression model for reservoir output weights learning.

\[W_{out} = YX^{T} \cdot (XX^{T} + \mathrm{_ridge} \times \mathrm{Id}_{_dim_in})\]

where \(W_out\) is the readout matrix learnt through this regression, \(X\) are the internal states, \(Y\) are the targets vectors, and \(_dim_in\) is the internal state dimension (number of units in the reservoir).

By default, ridge coefficient is set to \(0\), which is equivalent to a simple analytic resolution using pseudo-inverse.

Partial fit method allows to concurrently pre-compute \(XX^{T]\) and \(YX^{T}\) when several independent state sequences are provided, for performance, as suggested by [1].

Methods

`__init__`([ridge, workers, dtype])
`clean`()	Clean all internal parameters of the model.
`fit`([X, Y])	Fit states X to targets values Y following the model learning rule.
`initialize`([dim_in, dim_out])	Initialize the model internal parameters.
`partial_fit`(X, Y)	Partially fit the states X to the targets values Y.

Attributes

`dim_in`	Input dimension of the model (i.e. internal states dimension).
`dim_out`	Output dimension of the model.
`initialized`	A boolean indicating wether the internal parameters of the model are initialized or not.
`ridge`	Regularization coefficient of the model.
`Wout`

clean()[source]#: Clean all internal parameters of the model.

property dim_in#: Input dimension of the model (i.e. internal states dimension).

property dim_out#: Output dimension of the model.

fit(X=None, Y=None)[source]#

Fit states X to targets values Y following the model learning rule.

Parameters:

X (numpy.ndarray or list of numpy.ndarray) – Internal states of the reservoir.
Y (numpy.ndarray or list of numpy.ndarray) – Targets values for each states.

Returns:

A readout matrix of shape (targets dimension, states dimension + bias (=1)).

Return type:

numpy.ndarray

initialize(dim_in=None, dim_out=None)[source]#

Initialize the model internal parameters.

Parameters:

dim_in (int) – States dimension.
dim_out (int) – Targets dimension.

property initialized#: A boolean indicating wether the internal parameters of the model are initialized or not.

partial_fit(X, Y)[source]#

Partially fit the states X to the targets values Y. This method can be used to pre-compute some steps of the final fitting method.

Parameters:

X (numpy.ndarray or list of numpy.ndarray) – Internal states of the reservoir.
Y (numpy.ndarray or list of numpy.ndarray) – Targets values for each states.

property ridge#: Regularization coefficient of the model.