approximate_svd_linearity_transition

seas.ica.approximate_svd_linearity_transition(eig_val: numpy.ndarray)

Approximates the transition between the svd signal distribution and the noise floor.

Calculates the integral of the eigenvalue ‘influence’ per component, fits a 2 degree polynomial to the curve, and looks for the point at which the integrated eigenvalues first overshoot the polynomial fit. This transition point (multiplied by a hyperparameter) is used to inform the ICA n_components parameter.

Parameters

eig_val – The eigenvalues of the SVD decomposition.

Returns

The estimate of the SVD noise floor cutoff.

Return type

transition