def train(self):
"""Train HMMs"""
initial_hmm_files = self.get_initial_hmm_files()
if len(initial_hmm_files) == 0:
raise ModelException, "No initial HMM files found."
if not os.path.exists(self.TRAIN_HMM_ROOT):
os.makedirs(self.TRAIN_HMM_ROOT)
trainer = self.Trainer()
viterbi_trainer = ViterbiTrainer(self.ViterbiCalculator(),
non_diagonal=self.NON_DIAGONAL)
for file in initial_hmm_files:
char_code = int(os.path.basename(file).split(".")[0])
hmm = MultivariateHmm.from_file(file)
sset_file = os.path.join(self.TRAIN_FEATURES_ROOT,
str(char_code) + ".sset")
sset = self.get_sequence_set(sset_file)
output_file = os.path.join(self.TRAIN_HMM_ROOT,
"%d.xml" % char_code)
if self.TRAINING in (self.TRAINING_VITERBI, self.TRAINING_BOTH):
self.print_verbose("Viterbi training: " + output_file)
viterbi_trainer.train(hmm, sset)
if self.TRAINING in (self.TRAINING_BAUM_WELCH, self.TRAINING_BOTH):
self.print_verbose("Baum-Welch training: " + output_file)
trainer.train(hmm, sset)
hmm.write(output_file)
def init(self):
"""Init HMMs"""
self.load_char_dicts()
feature_files = self.get_train_feature_files()
if len(feature_files) == 0:
raise ModelException, "No feature files found."
if not os.path.exists(self.INIT_HMM_ROOT):
os.makedirs(self.INIT_HMM_ROOT)
for sset_file in feature_files:
char_code = int(os.path.basename(sset_file[:-5]))
sset = self.get_sequence_set(sset_file)
sset.char_code = char_code
hmm = self.get_initial_hmm(sset)
output_file = os.path.join(self.INIT_HMM_ROOT,
"%d.xml" % char_code)
self.print_verbose(output_file)
hmm.write(output_file)
def train(self):
"""Train HMMs"""
initial_hmm_files = self.get_initial_hmm_files()
if len(initial_hmm_files) == 0:
raise ModelException, "No initial HMM files found."
if not os.path.exists(self.TRAIN_HMM_ROOT):
os.makedirs(self.TRAIN_HMM_ROOT)
trainer = self.Trainer()
viterbi_trainer = ViterbiTrainer(self.ViterbiCalculator(),
non_diagonal=self.NON_DIAGONAL)
for file in initial_hmm_files:
char_code = int(os.path.basename(file).split(".")[0])
hmm = MultivariateHmm.from_file(file)
sset_file = os.path.join(self.TRAIN_FEATURES_ROOT,
str(char_code) + ".sset")
sset = self.get_sequence_set(sset_file)
output_file = os.path.join(self.TRAIN_HMM_ROOT,
"%d.xml" % char_code)
if self.TRAINING in (self.TRAINING_VITERBI, self.TRAINING_BOTH):
self.print_verbose("Viterbi training: " + output_file)
viterbi_trainer.train(hmm, sset)
if self.TRAINING in (self.TRAINING_BAUM_WELCH, self.TRAINING_BOTH):
self.print_verbose("Baum-Welch training: " + output_file)
trainer.train(hmm, sset)
hmm.write(output_file)
def init(self):
"""Init HMMs"""
self.load_char_dicts()
feature_files = self.get_train_feature_files()
if len(feature_files) == 0:
raise ModelException, "No feature files found."
if not os.path.exists(self.INIT_HMM_ROOT):
os.makedirs(self.INIT_HMM_ROOT)
for sset_file in feature_files:
char_code = int(os.path.basename(sset_file[:-5]))
sset = self.get_sequence_set(sset_file)
sset.char_code = char_code
hmm = self.get_initial_hmm(sset)
output_file = os.path.join(self.INIT_HMM_ROOT,
"%d.xml" % char_code)
self.print_verbose(output_file)
hmm.write(output_file)
def init(self):
self.load_char_dicts()
feature_files = self.get_train_feature_files()
if len(feature_files) == 0:
raise ModelException, "No feature files found."
if not os.path.exists(self.INIT_HMM_ROOT):
os.makedirs(self.INIT_HMM_ROOT)
ssets = []
# calculate the average number of observations for all characters
n_observations = 0
n_characters = 0
for sset_file in feature_files:
char_code = int(os.path.basename(sset_file[:-5]))
sset = self.get_sequence_set(sset_file)
sset.char_code = char_code
ssets.append(sset)
obs, chars = self.get_n_observations(sset)
n_observations += obs
n_characters += chars
avg_n_obs_per_char = float(n_observations) / n_characters
for sset in ssets:
hmm = self.get_initial_hmm(sset, avg_n_obs_per_char)
output_file = os.path.join(self.INIT_HMM_ROOT,
"%d.xml" % sset.char_code)
hmm.write(output_file)
def init(self):
self.load_char_dicts()
feature_files = self.get_train_feature_files()
if len(feature_files) == 0:
raise ModelException, "No feature files found."
if not os.path.exists(self.INIT_HMM_ROOT):
os.makedirs(self.INIT_HMM_ROOT)
ssets = []
# calculate the average number of observations for all characters
n_observations = 0
n_characters = 0
for sset_file in feature_files:
char_code = int(os.path.basename(sset_file[:-5]))
sset = self.get_sequence_set(sset_file)
sset.char_code = char_code
ssets.append(sset)
obs, chars = self.get_n_observations(sset)
n_observations += obs
n_characters += chars
avg_n_obs_per_char = float(n_observations) / n_characters
for sset in ssets:
hmm = self.get_initial_hmm(sset, avg_n_obs_per_char)
output_file = os.path.join(self.INIT_HMM_ROOT,
"%d.xml" % sset.char_code)
hmm.write(output_file)