fastIca - WeAreCAS

Q: What is the primary function of the fastIca action?

The fastIca action computes independent components using the fastICA method.

Q: How can I specify the nonquadratic function for negentropy approximation?

Use the gFunction parameter. It accepts "EXP" for the exponential function or "LOGCOSH" for the log-cosh function (which is the default).

Q: How do I configure the component extraction method?

You can configure it using the method parameter, which requires a 'name' subparameter set to "DEFLATION" or "SYMMETRIC". Optional subparameters include 'maxIter' (default 500) and 'tolerance' (default 0.0001).

Q: How can I suppress the centering and scaling of numeric variables?

To suppress centering, set the noCenter parameter to TRUE. To suppress scaling, set the noScale parameter to TRUE.

Q: What is the purpose of the eigenThreshold parameter?

The eigenThreshold parameter specifies the minimum proportion of variance an eigenvalue must explain to be included in the analysis; otherwise, it is discarded.

Q: How do I determine the number of independent components to be computed?

Use the n parameter. If you set n to 0 (the default value), the action computes the same number of independent components as there are numeric variables analyzed.

Description

Computes independent components by using the fastICA method. This action separates a multivariate signal into additive subcomponents assuming that the subcomponents are non-Gaussian and statistically independent. It includes options for data pre-processing (centering and scaling), different decorrelation methods (deflationary or symmetric), and nonquadratic functions for negentropy approximation.

Settings

Parameter	Description
table	Specifies the settings for an input table.
inputs	Specifies the numeric variables to be analyzed. If omitted, all numeric variables are used.
n	Specifies the number of independent components to be computed. If 0, it equals the number of numeric variables.
method	Specifies the settings for the component extraction method (maxIter, name, tolerance).
gFunction	Specifies the nonquadratic function to be used in the approximation of negentropy. Values: EXP, LOGCOSH.
seed	Specifies the seed value for pseudorandom number generation.
prefix	Specifies a prefix for naming the independent components in the output.
noCenter	When set to True, suppresses centering of the numeric variables to be analyzed.
noScale	When set to True, suppresses scaling of the numeric variables to be analyzed.
output	Specifies the output table to be created to contain observationwise statistics.

Data Preparation View data prep sheet

Create Mixed Signal Data

Create a dataset in CAS containing two mixed signals generated from a sine wave and a sawtooth wave.

Copied!

1
2	DATA mycas.signals;
3	DO t = 1 to 200;
4	s1 = sin(t/2);
5	s2 = mod(t, 10);
6	x1 = 0.6s1 + 0.4s2;
7	x2 = 0.4s1 + 0.6s2;
8	OUTPUT;
9	END;
10
11	RUN;
12

Examples

Perform Independent Component Analysis on the mixed signals variables x1 and x2.

SAS® / CAS Code Code awaiting community validation

Copied!

1
2	PROC CAS;
3	ica.fastIca / TABLE={name='signals'}, inputs={'x1', 'x2'}, seed=123;
4
5	RUN;
6

Result :
Generates the Mixing Matrix and Unmixing Matrix to separate the original signals.

Perform ICA using the symmetric method, the log-cosh function for negentropy, specifying 2 components, and saving the results.

SAS® / CAS Code Code awaiting community validation

Copied!

1
2	PROC CAS;
3	ica.fastIca / TABLE={name='signals'}, inputs={'x1', 'x2'}, n=2, method={name='SYMMETRIC', maxIter=1000}, gFunction='LOGCOSH', OUTPUT={casOut={name='ica_scores', replace=true}, component='Comp'}, seed=54321;
4
5	RUN;
6

Result :
Computes 2 components using symmetric decorrelation and saves the component scores to the 'ica_scores' table with prefix 'Comp'.

FAQ

What is the primary function of the fastIca action?

How can I specify the nonquadratic function for negentropy approximation?

How do I configure the component extraction method?

How can I suppress the centering and scaling of numeric variables?

What is the purpose of the eigenThreshold parameter?

How do I determine the number of independent components to be computed?

Table of Contents

Description

Create Mixed Signal Data

Examples

Basic FastICA Analysis

FastICA with Symmetric Decorrelation and Output

FAQ