ICA工具箱Infomax实现

ICA程序，利用Infomax算法实现，比较实用，建议跳过最优成分数估计，直接手动输入。

function [SAUllinfo]=icaML（XKpardebug_draw）

% icaML     : ICA by ML （Infomax） with square mixing matrix and no noise.

%

% function [SAUllinfo]=icaML（X[K][par]） Independent component analysis （ICA） using

%                                       maximum likelihood square mixing matrix and 

%                                       no noise [1] （Infomax）. Source prior is assumed 

%                                       to be p（s）=1/pi*exp（-ln（cosh（s）））. For optimization

%                                       the BFGS algorithm is used [2]. See code for 

%                                       references.

%                                       

%                                       X  : Mixed signals

%                                       K  : Number of source components.

%                                           For K=0 （default） number of sources are equal to number of

%                                           observations.

%                                           For K < number of observations SVD is used to reduce the

%                                           dimension.

%                                       par:  Vector with 4 elements:

%                                           （1）  :  Expected length of initial step

%                                           Stopping criteria:

%                                           （2）  :  Gradient  ||g||_inf <= par（2） 

%                                           （3）  :  Parameter changes ||dW||_2  <= par（3）*（par（3） + ||W||_2）

%                                           （4）  :  Maximum number of iterations 

%                                           Any illegal element in  opts  is replaced by its

%                                           default value  [1  1e-4*||g（x0）||_inf  1e-8  100]

%                                       debug_draw : Draw debug information

%                                       

%                                       S  : Estimated source signals with variance

%                                            scaled to one.

%                                       A  : Estimated mixing matrix

%                                       U  : Principal directions of preprocessing PCA. 
%                                            If K （the number of sources） is equal to the number
%                                            of observations then no PCA is performed and U=eye（K）.  
%                                       ll : Log likelihood for estimated sources

%                                       info :  Performance information vector with 6 elements:

%                                           （1:3）  : final values of [ll  ||g||_inf  ||dx||_2] 

%                                           （4:5）  : no. of iteration steps and evaluations of （llg）

%                                           （6）    : 1 means stopped by small gradient

%                                                    2 mea

 属性            大小     日期    时间   名称
----------- ---------  ---------- -----  ----
     文件       15496  2003-09-09 13:37  MLcorrected\icaML.m
     文件        3347  2006-06-22 14:15  MLcorrected\icaML_bic.m
     文件         845  2002-12-18 14:56  MLcorrected\script_demo.m