Why do I receive this error while classifying by GMM? (gmdistribution.fit)

Question

Negar 2015-10-29

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/251705-why-do-i-receive-this-error-while-classifying-by-gmm-gmdistribution-fit

I am trying to classify the data set of handwritten digits (0-9) by Gaussian mixture models, with 2 mixtures. The training data consists of 60000 examples of dimension 874 ( each example corresponding to 28X28 pixels, containing the gray level value of the image) . In the training phase, I try to fit the data to 10 classes, as follows:

clc
clear all
close all
load data_all
numMix = 2;
for i = 1:10
    % arange the training set corresponding to each class  
  x{i} =trainv(trainlab==(i-1),:); 
  a = x{i};
    % Initialize means, using Kmeans
    a(:, find(sum(abs(a)) == 0)) = [];
    [IDX,initMu] = kmeans(a,numMix);
    %  Initialize variance and mixture weights by arbitrary values.
    initCov =  eye(size(a,2));
    initVar= cat(3,initCov,initCov) ;
    initWeight = ones(1,2)/2;
    % Initial structure for the gmdistribution.fit function
    S.mu = initMu;
    S.Sigma = initVar;
    S.PComponents = initWeight;
    % Create models by implementing EM algorithm
    g{i} = gmdistribution.fit(a,numMix,'Start',S);
end
But when running the code, I receive this error:
Error using gmdistribution.fit (line 136)
The following column(s) of data are effectively constant: 1    2    3    4    5    6    7    8    9   10   11   12   13   14   15   16   17   18   19   20   21   22   23   24
25   26   27   28   29   30   31   32   33   34   35   36   37   38   39   43   47   48   49   50   51   52   53   54   55   56   57   58   59   62   66   67   77   78   79
80   81   82   83   84   85   86   87  111  112  113  114  115  139  140  141  142  143  167  168  169  170  171  172  197  198  199  226  251  252  253  254  279  280  281
282  307  308  309  311  336  337  364  365  392  393  394  395  419  420  421  422  423  447  448  449  450  451  475  477  478  479  503  505  506  507  531  532  533  559
560  561  588  589  590  617  618  619  645  646  647  671  672  673  674  675  676  699  700  701  702  703  704  705  706  720  721  722  723  724  725  726  727  728  729
730  731  732  733  734  735  736  737  738  739  740  741  742  743  744  745  746  747  748  749  750  751  752  753  754  755  756  757  758  759  760  761  762  763  764
765  766  767  768  769  770  771  772  773  774  775  776  777  778  779  780  781  782  783  784.
I tried to discard the zeros collumns. (Wondering if this would affect the final models?) I didn´t receive that error anymore, but this time I got the following error:
Error using gmcluster (line 180)
Ill-conditioned covariance created at iteration 2.

Error in gmdistribution.fit (line 174) [S,NlogL,optimInfo] =...

Error in digits_gmm (line 33) g{i} = gmdistribution.fit(a,numMix,'Start',S); I think this error has something to do with the way I initialize the covariance matrix. But couldn't figure out a better choice for that. Is there any workaround to this?