Opened 14 years ago

Closed 12 years ago

#269 closed request (fixed)


Reported by: Peter Owned by: Peter
Priority: major Milestone: yat 0.6
Component: classifier Version: trunk
Keywords: Cc:


I think this is a missing feature.

A MultiClassSVM is simply a wrapper around several SVMs. Everything should be straight forward. The only question is how to go from several classes to several binary classes. First choice is to only support one-versus-all (OVA). But there are other possibilities (such as all one vs one) and we should think about them when it is time.

Change History (6)

comment:1 Changed 13 years ago by Peter

Milestone: yat 0.x+yat 0.6
Status: newassigned

comment:2 Changed 12 years ago by Peter

(In [1855]) a first version of SvmMultiClass? (refs #269)

comment:3 Changed 12 years ago by Peter

Type: discussionrequest

SVM holds private variables C, max_epochs, and tolerance. These should be lumped into a class that can be held in both SVM and SvmMultiClass.

comment:4 Changed 12 years ago by Peter

(In [1861]) refs #269. Adding some simple tests. Also adding svm parameters. I chose not to bundle the params into a struct becuuse I don't see the point without deprecating a number (4) of functions in SVM or adding redundancy in interface. The downside of this is that it will be harder to maintain, but... the SVM part seems pretty stable.

comment:5 Changed 12 years ago by Peter

(In [1864]) adding test for train and predict. refs #269

comment:6 Changed 12 years ago by Peter

Resolution: fixed
Status: assignedclosed

(In [1865]) adding another test and fixing prediction when having empty training classes. closes #269

Note: See TracTickets for help on using tickets.