def get_phone_data(speeches): feats = [] for s in tqdm(speeches): f = featurizer.Featurizer(s) feat = f.get_phone_features() feats.append(np.array(feat)) return feats
def get_word_overlap_data(speeches): feats = [] for s in tqdm(speeches): f = featurizer.Featurizer(s) feat = [f.get_n_common_words()] feats.append(np.array(feat)) return feats
def get_substring_data(speeches): feats = [] for s in tqdm(speeches): f = featurizer.Featurizer(s) feat = f.get_common_substring_features() feats.append(np.array(feat)) return feats
def get_audio_data(speeches): feats = [] for s in tqdm(speeches): s.phrase_audio_features = s.get_phrase_audio_features() f = featurizer.Featurizer(s) feat = f.get_audio_features() feats.append(np.array(feat)) return feats
def get_mean_vector_data(speeches): feats = [] labs = [] for s in tqdm(speeches): f = featurizer.Featurizer(s) feats.append(f.get_mean_vector_features()) labs.append(f.get_binary_labels()) return feats, labs
def train(self, train_data):
f = featurizer.Featurizer(self.params)
print "SELECTING FEATURES"
f.select_features(train_data)
print "FEATURIZING TRAIN SET"
x = f.featurize(train_data)
y = utils.get_labels_from_post_list(train_data)
print "TRAINING MODEL"
self.model.fit(x, y)
self.f = f
print "TRAINING CLASSIFICATION REPORT"
test_classification = self.model.predict(x)
print classification_report(test_classification, y)
def traverse(l, path1, path2):
f = featurizer.Featurizer('vgg13')
worker = mp.current_process()
print('Traversing with ' + worker.name)
participants = os.listdir(path1)
#participants = ['p18', 'p19', 'p20', 'p21', 'p22', 'p23', 'p24']
for participant in participants:
part_path = os.path.join(path1, participant)
trials = os.listdir(part_path)
for trial in trials:
trial_path = os.path.join(part_path, trial)
l.acquire()
if is_locked(trial_path):
print(trial_path + ' was locked. Moving on.')
l.release()
continue
else:
print('Locking ' + trial_path)
lock(trial_path)
l.release()
feat_path = os.path.join(path2, participant, trial)
print('Working on ' + trial_path + ' with ' + worker.name)
os.system('mkdir -p ' + feat_path)
frames = get_frames(trial_path)
batches = get_batches(frames)
num = 1
while True:
try:
batch = batches.next()
except StopIteration:
break
print(worker.name + ' is doing a forward pass')
t1 = time.time()
batch_features = f.forward(ag.Variable(batch, volatile=True))
save_batch_features(feat_path, '{:05d}.t7'.format(num), batch_features)
print(worker.name + ' {:.2f}'.format(time.time()-t1) + ' seconds to complete forward pass')
num+=1
def get_vector_cosine_distance_data(speeches): feats = [] for s in tqdm(speeches): f = featurizer.Featurizer(s) feats.append(f.get_vector_cosine_distances()) return feats
def get_rst_data(speeches): feats = [] for s in tqdm(speeches): f = featurizer.Featurizer(s) feats.append(f.get_rst_features()) return feats
def get_binary_labels(speeches): labs = [] for s in tqdm(speeches): f = featurizer.Featurizer(s) labs.append(f.get_binary_labels()) return labs
def wrapper():
window, fts = feat_func()
return {
'WINDOW': window,
'FEATS': featurizer.Featurizer(types=fts, window=window),
}
