ExtendedInfomax

Namespace: RTPROCESSINGLIB  ·  Library: DSP Library

Python equivalent

mne.preprocessing.ICA (infomax) in MNE-Python.

#include <dsp/extended_infomax.h>

class UTILSLIB::ExtendedInfomax

Extended Infomax ICA (Lee et al., 1999).

Performs Independent Component Analysis using the extended infomax algorithm, which can separate both super-Gaussian and sub-Gaussian sources.

src/examples/ex_dsp_preprocessing/main.cpp

const MatrixXd centered = mixed.colwise() - mixed.rowwise().mean();                            // infomax expects zero-mean data
const InfomaxResult infomax = ExtendedInfomax::compute(centered, 3, 500, 0.1, 1e-7, true, 42); // the 0.001 default is too slow here

---

Static Methods

compute(matData, nComponents, maxIterations, learningRate, tolerance, extendedMode, seed)

Compute ICA decomposition using the extended infomax algorithm.

Parameters:

• matData : const Eigen::MatrixXd & Input data matrix (n_channels x n_times), should be mean-removed.

• nComponents : int Number of components to extract (-1 for n_channels).

• maxIterations : int Maximum number of iterations.

• learningRate : double Learning rate for weight updates.

• tolerance : double Convergence tolerance.

• extendedMode : bool If true, use extended mode (sub- and super-Gaussian).

• seed : unsigned int Random seed (0 for no seeding).

Returns:

• InfomaxResult — InfomaxResult containing unmixing/mixing matrices and sources.

---

Example

• Full source: src/examples/ex_dsp_preprocessing/main.cpp
• Build target: ex_dsp_preprocessing (runs under CTest)

Used in examples

ex_dsp_preprocessing

Preprocessing tools of the dsp library on synthetic data with a known ground truth.

src/examples/ex_dsp_preprocessing/main.cpp, line 135

const MatrixXd centered = mixed.colwise() - mixed.rowwise().mean();                            // infomax expects zero-mean data
const InfomaxResult infomax = ExtendedInfomax::compute(centered, 3, 500, 0.1, 1e-7, true, 42); // the 0.001 default is too slow here
ok &= expect(infomax.converged && worstRecovery(infomax.matSources, sources) > 0.99,
             QString("extended Infomax recovers the sub-Gaussian sources (worst |r| %1)").arg(worstRecovery(infomax.matSources, sources)));

Authors of this file

• Christoph Dinh <christoph.dinh@mne-cpp.org>