def predict(sentence: "Sentence string to predict output classes for"):
with open('features.json', 'r') as f:
feature_names = json.loads(f.read())
with open('labels.json', 'r') as f2:
labels = json.loads(f2.read())
core_nlp = CoreNLP()
tokens = core_nlp.tokenize_sentence(sentence)
features = extract_features(tokens)
X = pd.DataFrame(features).fillna(0) * 1
X = pd.get_dummies(X)
X = pd.DataFrame(X, columns = feature_names).fillna(0)
model = load_model('model.h5')
pred = model.predict(X)
result = pd.DataFrame(pred, columns=labels)
result = result.round(2)
result['word'] = [token['token'] for token in tokens]
result['vote'] = result[labels].idxmax(axis=1)
print(result)
return result
def train():
core_nlp = CoreNLP()
df = pd.read_table('data/train.txt', delimiter=' ', names=['word', 'label'])
sentences = " ".join(df['word']).replace(' ,', ',').replace(' .', '.').split('. ')
sentences = [s.strip().replace('.', '') + '.' for s in sentences]
features = []
for s in sentences:
tokens = core_nlp.tokenize_sentence(s)
features.append(extract_features(tokens))
X = pd.concat([pd.DataFrame(f) for f in features], sort=False)
X = X.fillna(0) * 1 # convert True to 0 before one_hot_enc
X = pd.get_dummies(X) # one_hot_enc
Y = pd.get_dummies(df['label']) # aka to_categorical
model = train_model(X, Y)
model.save('model.h5')
with open('labels.json', 'w') as f1: # Labels used
json.dump(Y.columns.tolist(), f1)
with open('features.json', 'w') as f2: # Features used
json.dump(X.columns.tolist(), f2)
return model
def init_features(info, f_extr_params):
exceptions = []
for track in info:
if not track['tags']:
exceptions.append(track['file_name'])
else:
track['features'] = f_extr_params['path_to'] + track[
'file_name'][:-4]
fe.extract_features(path_from=f_extr_params['path_from'],
path_to=f_extr_params['path_to'],
tmp_path_for_wavs=f_extr_params['tmp_path_for_wavs'],
exceptions=exceptions,
separator=f_extr_params['separator'],
methode=f_extr_params['methode'],
linear=f_extr_params['linear'])
return info
def feature_extraction(request):
filename = str(uuid.uuid1())
original_location = "Audio_Files/Uploaded_MP3/" + filename
with open(original_location, 'wb') as f:
f.write(request.body)
sound = AudioSegment.from_file(original_location)
location = "Audio_Files/WAV/" + filename + ".wav"
sound.export(location, format="wav")
if os.path.exists(original_location):
os.remove(original_location)
features = extract_features(location).reshape(1, -1)
if os.path.exists(location):
os.remove(location)
Voice_Disorder_Model = pickle.load(open("Voice_Disorder_Model.sav", 'rb'))
ans = Voice_Disorder_Model.predict(features)
print(int(ans[0]))
data = {'Disorder': int(ans[0])}
return JsonResponse(data)
def get_features(self, windows_size: int, with_previous_class_feature: bool = False) -> np.ndarray:
return fex.extract_features(self.__get_data_array(), windows_size,
with_previous_class_feature=with_previous_class_feature)[0]
train_data, dev_data, test_data, label_dictionary = data_fetching.fetch_data(
S_DATASET)
# PRE-PROCESSING
print('Data preprocessing...')
train_data['preprocessed'] = tweet_preprocessing.preprocess_data(
train_data['content'], 'main')
dev_data['preprocessed'] = tweet_preprocessing.preprocess_data(
dev_data['content'], 'main')
if B_TEST_PHASE is True:
test_data['preprocessed'] = tweet_preprocessing.preprocess_data(
test_data['content'], 'main')
# FEATURE EXTRACTION
print('Feature extraction...')
train_features = feature_extraction.extract_features(
train_data, train_data)
dev_features = feature_extraction.extract_features(
dev_data, train_data)
test_features = feature_extraction.extract_features(
test_data, train_data)
train_length = len(train_data)
test_length = len(test_data)
if B_TWEET_LENGTH_ANALYSIS:
utils.print_separator('Tweet length analysis')
print('Maximum lengths (words):')
print('Average lengths (words):')
print(' train: {}'.format(
train_features['tweet_length'].max()))
print(' test: {}'.format(test_features['tweet_length'].max()))
def main(s3_bucket_name, s3_object_key):
image = get_image(s3_bucket_name, s3_object_key)
image_features = extract_features(image)
print(image_features)