Hello,

my training set contains two classes.
I want to test classifier with test set, that contains just one instance. As a result I would like to get evaluated class lable (and posterior probability) of the test instance.

But when I use testc(D) I get an error message:

Could you please help me? How can I test just one instance?

Thanks,
Valeria

You can use labeld for finding labels and +(d*w*classc) for finding confidences.

Bob Duin

Thanks Bob,
it is exactly what I wanted!

But now I have an other question. Why if I train QDC, I get a complex numbers as a confidences?

You should not get it. It is the result of inverting (close to) singular matrices.
Some regularisation may hel (e.g. qdc([],1e-6)).

What is the date that is listed in your Contents.m file on the second line?

Bob

Hi Bob,
the date that is listed in my Contents.m file on the second line:

If I train TREEC or KNNC with the same data set I get normal numbers for the confidences. It happens only with QDC..

That is an old version, please download a new one from prtools.org

Bob

I downloaded a new version. Now I’ve got version 4.1.5 (30-Jul-2009).

But I still get either complex numbers or NaN values for confidence when I train QDC.
Is there some other ideas, why it happens?

Thanks,
Valeria

There is a newer version, please try it.

Bob Duin

Hello Bob,

now I’ve got the version 4.1.6 (15-Jan-2010).
But I stil get complex numbers with QDC and the warning:

Valeria

Can you send me the most simple version of the dataset and the qdc command you are using?

Please send it directly to me:

Bob Duin