def setUp(self):
super(TestPrototypeHMM,self).setUp()
self.p = PrototypeHMM(self.label,
num_states=0.5,
self_transprob=0.8,
next_transprob=0.2,
skip_transprob=1e-6)
self.p.train(self.ink_data, max_N=15)
def train(self, clustered_ink_data, verbose=False):
"""Train the classifier from clustered_data
Parameters
----------
clustered_ink_data : dictionary
clustered_ink_data[label] = [ [(ink,weight) from each cluster] ]
"""
self._trained_prototypes = []
for label in clustered_ink_data:
for ink_list in clustered_ink_data[label]:
if len(ink_list) > self.min_cluster_size:
ink_data, weights = zip(*ink_list)
proto = PrototypeHMM(label)
loglike = proto.train(ink_data, obs_weights=weights)
self._trained_prototypes.append(proto)
if verbose:
print ("Prototype for "
"%s (%d instances, avg_ll = %0.1f)"%(
label, len(ink_list), loglike/len(ink_list)))
self._compute_log_priors()
class TestPrototypeHMM(_BaseTest):
def setUp(self):
super(TestPrototypeHMM,self).setUp()
self.p = PrototypeHMM(self.label,
num_states=0.5,
self_transprob=0.8,
next_transprob=0.2,
skip_transprob=1e-6)
self.p.train(self.ink_data, max_N=15)
def test_serialization(self):
p_data = self.p.toJSON()
q = PrototypeHMM(None)
q.fromJSON(p_data)
self.assertEqual(p_data, q.toJSON())
def test_score(self):
score0 = self.p.score(self.ink_data[0])
self.assertAlmostEqual(score0, 47.772, delta=1e-3)
score1 = self.p.score(self.ink_data[1])
self.assertAlmostEqual(score1, 53.725, delta=1e-3)
def _train(algorithm, label, obs, obs_weights):
if algorithm == 'hmm':
prot = PrototypeHMM(label)
prot.train(obs, obs_weights=obs_weights)
else:
prot = PrototypeDTW(label)
prot.train(obs, obs_weights=obs_weights,
center_type='centroid')
return prot
def test_serialization(self):
p_data = self.p.toJSON()
q = PrototypeHMM(None)
q.fromJSON(p_data)
self.assertEqual(p_data, q.toJSON())