Question

Is there a way of specifying multiple labels for an example in the Vowpal Wabbit data format?

I tried comma's and it doesn't complain, but I am not sure if it does what I expect...

So e.g:

2,3 | f1:3 f3:23 
2 | f1:2 f3:34

The first line is an example that is both in class 2 and in class 3.

I do see some discussion on the vowpal mailing list but I am not sure if anything has happened since.

Tom

Was it helpful?

Solution

It looks like you should use CSOAA which allows multiple labels. They need to be natural numbers and less than or equal to K where K is the number of classes. Like this:

1 2 3 class1 | f1:2 f2:3
4 2 5 class2 | f3:7 f5:3
2 3 class3 | f4:2 f2:2

For more detail, see the Vowpal Wabbit documentation: https://github.com/JohnLangford/vowpal_wabbit/wiki/Cost-Sensitive-One-Against-All-%28csoaa%29-multi-class-example

Licensed under: CC-BY-SA with attribution
Not affiliated with StackOverflow
scroll top