Hi,
I'm looking for software for tagging of time series data using Support
Vector Machines, i.e. a (freely downloadable, in source-code form) package
that goes beyond simple classification/regression task in isolation in
that it incorporates a trellis-based dynamic programming search over
sequences (e.g., in language processing: a sentence = sequence of words)
and identifies the most likely path (label sequence).
Standard packages like Weka, Torch/SVMTorch, SVMlight etc either do not
provide such a search (they only look at comma-separated _individual_
feature vectors and their category labels without a notion of an
underlying optimal time sequence search). Others, like SVMTool, have
currently have the feature set hardwired (Part-of-Speech tagging only),
which renders them less reusable for other tasks (although there seems a
new version in the making).
So far I have discovered Yamcha <http://chasen.org/~taku/software/yamcha/>;
are there any other free implementations out there?
Thanks in advance & regards,
Jochen
--
Jochen L Leidner <[log in to unmask]>
ICCS <[log in to unmask]>
University of Edinburgh <http://www.iccs.informatics.ed.ac.uk/~s0239229/>
|