Closed franz101 closed 1 year ago
It's fairly difficult to interpret the influence of sub-sequences on classification for both KNNClassifier
and DeepGRU
.
However with HMMClassifier
+ GMMHMM
, you can inspect the state emission distributions and transition matrices for the HMMs belonging to each class. This will give you some idea about how each HMM is learning to model the sequences.
Note that HMMClassifier
on v0.12.1/master is based on a version of hmmlearn
that had major bugs in it. I suggest installing from the dev
branch (which has been updated) until I make the next release.
Is there a best practise for understanding which sub-sequences belong to which classes? Or what factors the model made decide?