The pooled covariance matrix of TRAINING must be positive definite.

clc
clear all

load featurs_T
load featurs_S
load Group_Train
load Group_Test

cv_x=cov(Feat1);
[V,D] = eig(cv_x);
d=diag(D);
d=d(end:-1:1);
sm_d=cumsum(d) /sum(d);
idx=find(sm_d>0.99);
T=[V(:,end:-1:idx(1))]';
new_feat1=T*Feat1';
%TrainingSet= new_feat1';
new_feat2=T*Feat2';
%TestSet= new_feat2';

TrainingSet = new_feat1'; 
TestSet = new_feat2';
Group_Train1 = Group_Train1';
Group_Test1 = Group_Test1';


%------------------------


% result1= multisvm(TrainingSet,Group_Train1,TestSet,Group_Test1);
result1= classify(TestSet,TrainingSet,Group_Train1,'linear');

testresult = result1;

Accuracy = mean(Group_Test1==result) * 100;
fprintf('Accuracy = %.2f\n', Accuracy);
fprintf('error rate = %.2f\n ', mean(result ~= Group_Test1 ) * 100);

Error using classify (line 233)

The pooled covariance matrix of TRAINING must be positive definite.

Error in HOG2 (line 31)

result1= classify(TestSet,TrainingSet,Group_Train1,'linear');

[yanqi liu]

yes，sir，may be upload your mat file to debug

[sun rise]

Thank you in advance

These are my files

[yanqi liu]

yes，i tried，the data is sparse，may be use pca to reduce dimension

[sun rise]

I actually used pca to reduce dimensions. But the error is still there

[sun rise]

Feat1 = pca(Feat1);
Feat2 = pca (Feat2);

But why is pca decreasing the number of images and thus I get this error

The length of GROUP must equal the number of rows in TRAINING.

This is evident in the workspace

Answer 1

[Matt J]

I suggest you calculate the pooled covariance matrix and verify whether the error message is accurate.

[sun rise]

https://www.mathworks.com/matlabcentral/answers/1633570-the-pooled-covariance-matrix-of-training-must-be-positive-definite#comment_1961110