Why I only get one column of probability estimates for two classes in libsvm?

Question

I copied the code from Multi-class classification in libsvm to get the probability estimates for each class.

However, I have an error in that my probability estimates p only has one column when there should be two columns. I checked my model and it states there are two classes (model.nr_class = 2) and (model.Label = [0;1]).

Can someone explain?

My probability estimates range from -0.35 to 1.2057 so they are not between 0 and 1. Giving the option -b 0 and -b 1 returns the same result.

if there are two classes and you have the probability of class A, isn't p(B|x) just 1 - p(A|x)? — Ben Allison, Mar 28 '14 at 13:25
Oh, I figured it out. You have to give -b 1 to both training and testing to get two columns. Thanks! — Anonymous, Mar 31 '14 at 17:55

score 0 · Answer 1 · answered Mar 28 '14 at 01:56

0

I guess you miss the -b option in training. Check libsvm documentation.

answered Mar 28 '14 at 01:56

kyamagu

551
2
5

score 0 · Answer 2 · answered Mar 30 '14 at 03:04

0

as you are having 2 classes, its clear that if the prob of class A is 0.xxxx then prob of another class is 1-0.xxxx, that you can clearly calculate...

answered Mar 30 '14 at 03:04

kAmol

1,297
1
18
29

Why I only get one column of probability estimates for two classes in libsvm?

2 Answers2