Hi Dan. Scikit-learn focuses on "flat" signals and algorithms and we don't usually add algorithms on time-series or nd-data, as that would significantly widen the scope and complicate API. Maybe we should add this to the FAQ.
FWIW I didn't have a very good experience when working with convolutional (shouldn't it be that?) NMF. Why no use an autoencoder approach? Andy On 04/09/2015 08:20 AM, Dan Stowell wrote: > Hi all, > > Does anyone here have any experience/tips for _convolutive_ NMF in > scikit-learn (or in numpy more generally)? scikit-learn has NMF > decomposition, hooray, but nothing for the convolutive version. > t > "Convolutive" here means that the bases are not just 1-dimensional but > 2-dimensional: the basic NMF model X=WH is expanded so that each element > of H is convolved with an element of W, not just multiplied. Useful for > timeseries such as audio spectrograms: > > http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.320.5545&rep=rep1&type=pdf > > http://eprints.maynoothuniversity.ie/1375/1/getPDF2.pdf > > Thanks > Dan > ------------------------------------------------------------------------------ BPM Camp - Free Virtual Workshop May 6th at 10am PDT/1PM EDT Develop your own process in accordance with the BPMN 2 standard Learn Process modeling best practices with Bonita BPM through live exercises http://www.bonitasoft.com/be-part-of-it/events/bpm-camp-virtual- event?utm_ source=Sourceforge_BPM_Camp_5_6_15&utm_medium=email&utm_campaign=VA_SF _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
