Is there an easy way to calculate multiclass precision, recall and F-Measure validation?

[haifengl]

Is there an easy way to calculate multiclass precision, recall and F-Measure validation?

[haifengl]

Check out the package smile.validation. We have all of them (and something else) there.

[owlmsj]

Sure, but Precision, Recall and FMeasure implementation is only for one-class validation, right? It test 0 (false) and 1 (true), not the class-number. Or what am I doing wrong?

[owlmsj]

This issue is like this question : http://stats.stackexchange.com/questions/51296/how-to-calculate-precision-and-recall-for-multiclass-classification-using-confus

[haifengl]

Ok. Let's create a class to calculate confusion matrix and then update these measure classes for multi-class.

[owlmsj]

Hi, I did that in meantime. Any suggestions about this new class name?

[haifengl]

ConfusionMatrix should be fine. You can make a pull request and we can start from there. Thanks!

[owlmsj]

Okay! Just a little con: Project 'Smile' is missing required Java project: 'SmileTestData' . Where can I find this project?

[haifengl]

SmileTestData contains various datasets that I download from internet. As I am not the owner of these datasets, I cannot include them in github.

[owlmsj]

Ah, no problem. I just remove the dependency and it was OK!

[haifengl]

The maven .pom files have no such dependency. Only for eclipse project files. BTW, SmileTestData is only for unit tests. You don't need them to compile the packages. Thanks!

[haifengl]

Thanks for the great work! The ConfusionMatrix is merged into master.