changing the cutoff

The front end.
Post Reply
liid
Posts: 15
Joined: Tue Jun 19, 2018 5:19 am

changing the cutoff

Post by liid » Tue Jun 26, 2018 10:48 pm

Hi,

I am doing a 3-class classification problem.
From the ROC I figured that I need to change the cutoffs. How can I do that in the software?

In general, what are the cutoffs the software uses for a 3-class classification?

Thanks!

marek [BayesFusion]
Site Admin
Posts: 270
Joined: Tue Dec 11, 2007 4:24 pm

Re: changing the cutoff

Post by marek [BayesFusion] » Wed Jun 27, 2018 5:52 am

When validating a model, GeNIe picks for each record the most likely class, so while with two classes the implicit cut-off is 0.5, the implicit cut-off for three classes is 1/3. There is no way you can pick your own cut-off when running validation in GeNIe. What you can do in a simple way is generate an output file, which will contain the input records and probability distribution over the classes. You can open the output file in Excel and then use your own threshold (you can accomplish this by adding one more column with a function that picks a class). Alternatively, you can program your own criterion for picking a class when you use SMILE for inference.
I hope this helps,

Marek

Post Reply