UTILSLIB::PicardIca Class Reference

Independent Component Analysis using the Picard algorithm. More...

#include <picard_ica.h>

Static Public Member Functions
static IcaResult	run (const Eigen::MatrixXd &matData, int nComponents=-1, int maxIter=200, double tol=1e-7, int lbfgsMemory=7, int randomSeed=42)

Detailed Description

Independent Component Analysis using the Picard algorithm.

Picard uses a preconditioned L-BFGS strategy to minimise the mutual information of the components. Compared to FastICA it typically converges in fewer iterations and handles badly conditioned data more robustly.

IcaResult res = PicardIca::run(matRawData, 20);
QVector<int> exclude = {0, 3};
MatrixXd clean = ICA::excludeComponents(matRawData, res, exclude);

Definition at line 60 of file picard_ica.h.

Member Function Documentation

run()

IcaResult PicardIca::run ( const Eigen::MatrixXd & matData, int nComponents = -1, int maxIter = 200, double tol = 1e-7, int lbfgsMemory = 7, int randomSeed = 42 )

static

Run the Picard ICA algorithm.

Parameters

[in]	matData	Input data (n_channels x n_samples).
[in]	nComponents	Number of components to extract (-1 = all channels).
[in]	maxIter	Maximum number of iterations (default 200).
[in]	tol	Convergence tolerance (default 1e-7).
[in]	lbfgsMemory	L-BFGS memory length (default 7).
[in]	randomSeed	Seed for initial weight randomisation (default 42).

Returns

IcaResult with mixing/unmixing matrices and source time series.

Definition at line 70 of file picard_ica.cpp.

---

The documentation for this class was generated from the following files:

• src/libraries/dsp/picard_ica.h
• src/libraries/dsp/picard_ica.cpp