def get_combined_data():
# reading train data
train = read_training_data()
# reading test data
test = read_test_data()
# import ipdb; ipdb.set_trace()
# extracting and then removing the targets from the training data
targets = train.Survived
train.drop('Survived', 1, inplace=True)
# merging train data and test data for future feature engineering
# import ipdb; ipdb.set_trace()
combined = train.append(test)
combined.reset_index(inplace=True)
combined.drop('index', inplace=True, axis=1)
return combined
def predict(self):
X_test = read_data.read_test_data()
return self.model.predict(X_test, verbose=1)
sp_train = params.sp_train
sp_cont = params.sp_cont
ap_train = params.ap_train
ap_cont = params.ap_cont
spectral_data, aperiodic_data, label_data, cutoff_points = read_training_data(
model_name, load=load_data)
singing_model = SingingModel(spectral_data, aperiodic_data, label_data,
cutoff_points, model_name)
if sp_train:
singing_model.train_model(SPECTRAL_MODE, sp_cont)
if ap_train:
singing_model.train_model(APERIODIC_MODE, ap_cont)
label_data, frequency = read_test_data(model_name)
spectral_output = singing_model.inference(label_data, SPECTRAL_MODE)
aperiodic_output = singing_model.inference(label_data, APERIODIC_MODE,
spectral_output)
spectral_output, aperiodic_output = decode_envelopes(
spectral_output, aperiodic_output, params.sample_rate, model_name)
construct_audio(spectral_output, aperiodic_output, frequency, output_name)
def test_performance(model):
test_x, test_y = read_test_data()
test_pred = model.predict(test_x, batch_size=512, verbose=1)
evals = get_evals(test_y, test_pred)
return evals
import tensorflow as tf
from tensorflow._api.v1.keras import layers
import read_data
x_train, y_train = read_data.read_train_data()
x_test, y_test = read_data.read_test_data()
num, H, W, _ = x_train.shape
model = tf.keras.models.Sequential([
tf.keras.layers.Conv2D(filters=16,
kernel_size=3,
strides=(1, 1),
padding='valid',
data_format='channels_last',
activation='relu',
use_bias=True,
input_shape=(H, W, 1)),
tf.keras.layers.Conv2D(filters=32,
kernel_size=3,
strides=(1, 1),
padding='same',
data_format='channels_last',
activation='relu',
use_bias=True,
input_shape=(H - 2, W - 2, 16)),
tf.keras.layers.Conv2D(filters=64,
kernel_size=3,
strides=(1, 1),
padding='same',
data_format='channels_last',
activation='relu',
if __name__ == "__main__":
TRAIN_DATA, LEMMA2SENSES, LEMMA2INT = read_data.read_train_data(
read_data.read_x("ALL.data.xml")[0],
read_data.read_y("ALL.gold.key.bnids.txt"), True)
MAX_NB_SENSES = max([len(LEMMA2SENSES[k]) for k in LEMMA2SENSES
]) # max number of senses among all target words
MAX_NB_TARGETS = len(LEMMA2SENSES) # how many target words
# load word embedding initialized by init_emb (run init_emb first if you don't have this file)
with open('pretrained_vectors/needed' + '.pkl', 'rb') as f:
WORD_VECTORS = pickle.load(f)
WORD_VECTORS["_drop_"] = np.random.uniform(
-0.1, 0.1, 300) # add drop vector for drop words
NB_EPOCHS = 100 # number of epochs to train
x_val, y_val, _ = read_data.read_test_data(
LEMMA2INT, LEMMA2SENSES, WORD_VECTORS) # read validation data
"""train models"""
tf.reset_default_graph()
train_model = models.Model2(MAX_NB_SENSES, 32, MAX_NB_TARGETS)
val_model = models.Model2(MAX_NB_SENSES,
32,
MAX_NB_TARGETS,
is_training=False)
print("train models created")
"""run train models"""
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
VAL_ACC_LIST = []
LOSS_LIST = []
TRAIN_ACC_LIST = []