check pomegranate

check_pomegranate.py

from pomegranate import *
from pomegranate.parallel import log_probability
from pomegranate.io import SequenceGenerator
from pomegranate.kmeans import Kmeans
from pomegranate.distributions.IndependentComponentsDistribution import IndependentComponentsDistribution
from pomegranate.kmeans import Kmeans
from pomegranate.io import SequenceGenerator
import numpy 

#get code from pomegranate lib
def get_group_with_kmeans(X,n_components):
	X_, labels_ = [], []

	data_generator = SequenceGenerator(X)
	initialization_batch_size = len(data_generator)

	data = data_generator.batches()
	for i in range(initialization_batch_size):
	    batch = next(data)
	    X_.extend(batch[0])
	        

	X_concat = numpy.concatenate(X_)

	if X_concat.ndim == 1:
	    X_concat = X_concat.reshape(X_concat.shape[0], 1)

	n, d = X_concat.shape

	clf = Kmeans(n_components)
	clf.fit(X_concat)
	y = clf.predict(X_concat)

	groups = []
	for i in range(n_components):
		groups.append(X_concat[y == i])

	return groups

def main():

	X =  numpy.array([
		[43.50],[11.73],[11.78],[11.51],[11.37],
		[22.10],[21.73],[21.78],[21.51],[21.37],
		[32.10],[31.73],[31.78],[31.51],[31.37],
		[42.10],[41.73],[41.78],[41.51],[41.37]
		])

	n_components = 2
	number_groups = 4

	groups = get_group_with_kmeans(X, number_groups)

	mixtures = []
	for i in range(len(groups)):
		mixtures.append(NormalDistribution(numpy.mean(groups[i]), numpy.std(groups[i]), True))

	#Define 2 mixtures for each group
	dists = []
	for i in range(n_components):
		dists.append(GeneralMixtureModel([mixtures[len(dists)*2],mixtures[len(dists)*2+1]]))
	starts = numpy.ones(n_components, dtype='float32') / n_components
	transitions = numpy.ones((n_components, n_components), dtype='float32') / n_components

	model = HiddenMarkovModel.from_matrix(transitions, dists, starts)

	model.fit(X)
	

	observation=[43.50]
	trans, ems = model.forward_backward(observation)
	#sum row from emission table shoul be equal 1 -- OK
	print(numpy.exp(ems[0]))

	#Calculate sum probabilities for each state, shoul be equals 1 - NOt OK
	log_probs = []
	for j in range(len(model.states)):
		if(model.states[j].distribution != None): #skip silent states
			for m in range(len(model.states[j].distribution.distributions)):
				weight = numpy.exp(model.states[j].distribution.weights[m])
				log_prob = model.states[j].distribution.distributions[m].log_probability(observation)
				log_probs.append(numpy.exp(log_prob) * weight)
	print(sum(log_probs))

main()