Search All of the Math Forum:

Views expressed in these public forums are not endorsed by NCTM or The Math Forum.

Notice: We are no longer accepting new posts, but the forums will continue to be readable.

Topic: Counting several predictions as positive
Replies: 0

 Jay Posts: 31 Registered: 7/23/11
Counting several predictions as positive
Posted: Jun 19, 2013 4:34 AM

Hi!

I want to evaluate a classifier that predicts the orientation of an object. I split the possible orientations into 8 classes (F (=Front), FL (= Front Left), L (= Left), BL (= Back Left), B (=Back), BR = (Back Right) , R (=Right), FR = (Front Right)). You can imagine a circle being split clockwise by these classes, starting at 6 pm with Front.

As a first rough evaluation I want to say that an orientation was correctly classified when either the orientation was precisely predicted or the classifier predicted one of the two neighboring orientations.
E.g.: An example is truly left, now the classifier could predict FL, L and BL to be correct.

I hope this was understandable so far...

I thought I could achieve this behavior by using something like:
cpLeft = classperf(orientationLabels, 'Positive', {'frontleft','left', 'backleft'}, 'Negative', {'back','backright', 'right', 'frontright', 'front'});

However, I believe this is not what I want. Because here examples are counted as true positives, where the classifier e.g. predicted FL and the true class was BL.

Was it understandable what I want to achieve? Can I achieve what I want with classperf? If not, are there predefined ways how I could do it?

Thanks!