hello every body
always we use labeld dataset for traine a classifier then use crossval for error calculation, for classify of another dataset we can use this trained classifier,but how we can recognize error of this classified dataset and accuracy of each class ?

thankyou

Use testd, confmat or the second output parameter of testc to find errors per class.

Bob Duin

hello
by ‘confmat’ i write :

‘a’ : labeld dataset

[tr,test]=gendat(a,.3);
[d,w]=emclust(tr,nmc,6);
s=a*w;
confmat(tr)

OR

J1=emclust(tr,nmc,6);
J2=emclust(tr,nmc,6);
[J3,c]=matchlab(J1,J2);
confmat(J1,J3);

is it right?

another qustion : for new unlabeld dataset (’b’) that we want to classify by w how earn error?

x=b*w;

ERROR?????

The first example is definitely wrong, by

confmat(tr)

you will get an error as two label lists are expected. Use

confmat(getlabels(tr,d);

for the trainset or

confmat(getlabels(test),test*w*labeld)

for the testset or

confmat(s)

for the classification of a. The second example is fine. It shows the difference
between two clustering.

The classification error is defined as the difference between true and estimated labels.
If there are no true labels, a classification error cannot be computed.

Bob Duin

how can i display missclassificatiom images
from tested_dataset file

i have

tested_file = anewn

oreginal images folder

Suppose we have a dataset A derived from images. So A(n,:) contains the
features of image n.
During construction of A objects receive an identifier wich can be inspected
by getident(A,’ident’);
Suppose it has been split in a training set and a testset, a classifier is
constructed and the testset is classified:

[trainset,testset] = gendat(A,0.5);
W = trainset*fisherc;
elab = testset*W*labeld;

elab contains the estimated labels. Let’s find where they are incorrect, i.e.
different from the true labels of the testset.

N = find(~isequal(elab,getlabels(testset)));

In N the indices of the incorrectly classified testobjects are stored. If we find
the coressponding object identifiers, stored in testset, we can trace this back to
A and images:

Ntestset_ident = getident(testset,N);
J = findident(A,Ntestset_ident);

J contains the indices in A of the erroneously classified testobject generated from A.
Depending on how A was constructed from the images, J points thereby to the images that
should be displayed. (If A is constructed by a dataset command J equals Ntestset_ident,
so the last statement might not be needed).

Best regards,

Bob Duin
Now we have to find to which objects in A these

when i classify my image by a classifier ,i reach to error for example 8% but picture of classified images isnot good, what does it mean?

What does it mean that the ‘image is not good’? Are you classifying pixels or images?

Bob Duin