[SciPy-user] clustering multivariate timeseries

mani sabri mani.sabri@gmail....
Wed Jan 23 04:01:01 CST 2008


Hi 
Sorry for irrelevant subject.I'm learning data mining with incremental
experiments approach and I ended up clustering my sensors (multivariate
timeseries)with wavelet 2d transform(pywt)(compressing the data to a single
timeseries and using dynamic time wrapping as distance measurement and
passing distance matrix to hierarcial clustering algorithm(orange HCA)) but
I ended up with a cluster of frames in which sensor A is similar to sensor
B! and sensor B is similar to sensor C and so on (far from ideal: sensor A
being similar to sensor A and sensor B being similar to sensor B and ...)I'm
not surprise because AFAIK 2d wavelet transform looks at data as a 2d
picture not different timeseries.
How can I solve this problem? Do hmms (hidden markov models) solve this
problem? Is there any module available in numpy context?

I really appreciate your suggestion.
Regards, Mani





More information about the SciPy-user mailing list