def create_hmm():
emit_states = range( len(vq.gen_alphabet()) )
emit_domain = Alphabet( emit_states )
trans_p = [ [0.9, 0.05 , 0.05],
[ 0.1, 0.8 ,0.1],
[ 0.1, 0.1, 0.8]]
emit_p = generate_emit_p()
pi = [ 0.9, 0.05, 0.05 ]
return HMMFromMatrices( emit_domain,
DiscreteDistribution(emit_domain),
trans_p,
emit_p,
pi
),emit_domain
def generate_emit_p():
straight = [ vq.dataset_to_alphabet(gen.straight()) for x in xrange(50) ]
left_turns = [ vq.dataset_to_alphabet(gen.left_turn()) for x in xrange(100) ]
right_turns = [ vq.dataset_to_alphabet(gen.right_turn()) for x in xrange(100) ]
data_sets = [ straight, left_turns, right_turns ]
gen_data_set = { 0: {}, 1: {}, 2: {} }
for idx, dataset in enumerate(data_sets):
for sample in dataset:
for alpha in sample:
if alpha not in gen_data_set[idx]:
gen_data_set[idx][alpha] = 1
else:
gen_data_set[idx][alpha] += 1
alpha_size = len(vq.gen_alphabet() )
emit_p = [ [0.01 for x in xrange(alpha_size) ] for x in xrange(3) ]
for idx, stats in gen_data_set.iteritems():
for k,v in stats.iteritems():
emit_p[idx][k] = v
return normalize_emit_probabilites( emit_p )