def train(x_train, y_train, x_test, y_test):
# Init Keras Model here
model = VDCNN(num_classes=y_train.shape[1],
depth=FLAGS.depth,
sequence_length=FLAGS.sequence_length,
shortcut=FLAGS.shortcut,
pool_type=FLAGS.pool_type,
sorted=FLAGS.sorted,
use_bias=FLAGS.use_bias)
model.load_weights('./checkpoints/vdcnn_weights_val_acc_0.6634.h5',
by_name=True)
model.compile(optimizer=SGD(lr=0.001, momentum=0.9),
loss='categorical_crossentropy',
metrics=['accuracy'])
model_json = model.to_json()
with open("vdcnn_model.json", "w") as json_file:
json_file.write(model_json) # Save model architecture
time_str = datetime.datetime.now().isoformat()
print("{}: Model saved as json.".format(time_str))
print("")
# Trainer
# Tensorboard and extra callback to support steps history
tensorboard = TensorBoard(log_dir='./logs',
histogram_freq=50,
batch_size=FLAGS.batch_size,
write_graph=True,
write_images=True)
checkpointer = ModelCheckpoint(
filepath="/checkpoints/vdcnn_weights_val_acc_0.5931.h5",
period=1,
verbose=1,
save_best_only=True,
mode='max',
monitor='val_acc')
loss_history = custom_callbacks.loss_history(model, tensorboard)
evaluate_step = custom_callbacks.evaluate_step(model, checkpointer,
tensorboard,
FLAGS.evaluate_every,
FLAGS.batch_size, x_test,
y_test)
# Fit model
model.fit(
x_train,
y_train,
batch_size=FLAGS.batch_size,
epochs=FLAGS.num_epochs,
validation_data=(x_test, y_test),
verbose=1,
callbacks=[checkpointer, tensorboard, loss_history, evaluate_step])
print('-' * 30)
time_str = datetime.datetime.now().isoformat()
print("{}: Done training.".format(time_str))
K.clear_session()
print('-' * 30)
print()
def train(x_train, y_train, x_test, y_test):
# Init Keras Model here
x_train, x_val, y_train, y_val = train_test_split(x_train,
y_train,
test_size=0.2,
random_state=1)
model = VDCNN(num_classes=1,
depth=FLAGS.depth,
sequence_length=FLAGS.sequence_length,
shortcut=FLAGS.shortcut,
pool_type=FLAGS.pool_type,
sorted=FLAGS.sorted,
use_bias=FLAGS.use_bias)
adam = Adam(lr=0.0001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(optimizer=adam,
loss='binary_crossentropy',
metrics=['accuracy'])
model_json = model.to_json()
with open("vdcnn_model.json", "w") as json_file:
json_file.write(model_json) # Save model architecture
time_str = datetime.datetime.now().isoformat()
print("{}: Model saved as json.".format(time_str))
print("")
# Trainer
# Tensorboard and extra callback to support steps history
tensorboard = TensorBoard(log_dir='./logs',
histogram_freq=50,
batch_size=FLAGS.batch_size,
write_graph=True,
write_images=True)
checkpointer = ModelCheckpoint(
filepath=
"/home/nujud/Desktop/VDCNN-keras_version (1)/logs/checkpoints/vdcnn_weights_val_acc_{val_acc:.4f}.h5",
period=1,
verbose=1,
save_best_only=True,
mode='max',
monitor='val_acc')
x_train, x_val, y_train, y_val = train_test_split(x_train,
y_train,
test_size=0.3,
random_state=1)
loss_history = custom_callbacks.loss_history(model, tensorboard)
evaluate_step = custom_callbacks.evaluate_step(model, checkpointer,
tensorboard,
FLAGS.evaluate_every,
FLAGS.batch_size, x_val,
y_val)
testcallback = custom_callbacks.TestCallback(model, (x_test, y_test))
# Fit model
model.fit(x_train,
y_train,
batch_size=FLAGS.batch_size,
epochs=FLAGS.num_epochs,
validation_data=(x_val, y_val),
shuffle=True,
verbose=1,
callbacks=[
checkpointer, tensorboard, loss_history, evaluate_step,
testcallback
])
print('-' * 30)
time_str = datetime.datetime.now().isoformat()
print("{}: Done training.".format(time_str))
K.clear_session()
print('-' * 30)
print()
def train(x_train, y_train, x_test, y_test):
# Init Keras Model here
# class_weights = class_weight.compute_class_weight('balanced',
# np.unique(y_train),
# y_train)
class_weight_dict = {0: 0.7, 1: 1.6}
# class_weight_dict = dict(enumerate(class_weights))
model = VDCNN(
num_classes=y_train.shape[1], #
depth=FLAGS.depth,
sequence_length=FLAGS.sequence_length,
shortcut=FLAGS.shortcut,
pool_type=FLAGS.pool_type,
sorted=FLAGS.sorted,
use_bias=FLAGS.use_bias)
model.compile(optimizer=SGD(lr=0.01, momentum=0.9),
loss='binary_crossentropy',
metrics=['accuracy'])
model_json = model.to_json()
with open("vdcnn_model.json", "w") as json_file:
json_file.write(model_json) # Save model architecture
time_str = datetime.datetime.now().isoformat()
print("{}: Model saved as json.".format(time_str))
print("")
# Trainer
# Tensorboard and extra callback to support steps history
tensorboard = TensorBoard(log_dir='./logs',
histogram_freq=50,
batch_size=FLAGS.batch_size,
write_graph=True,
write_images=True)
checkpointer = ModelCheckpoint(
filepath="./checkpoints/vdcnn_weights_val_acc_{val_acc:.4f}.h5",
period=1,
verbose=1,
save_best_only=True,
mode='max',
monitor='val_acc')
loss_history = custom_callbacks.loss_history(model, tensorboard)
evaluate_step = custom_callbacks.evaluate_step(model, checkpointer,
tensorboard,
FLAGS.evaluate_every,
FLAGS.batch_size, x_test,
y_test)
clr = CyclicLR(base_lr=0.004,
max_lr=0.01,
step_size=2000.,
mode='triangular',
gamma=0.99994)
validation_data = (x_test, y_test)
metrics = Metrics(validation_data)
# Fit model
model.fit(x_train,
y_train,
batch_size=FLAGS.batch_size,
epochs=FLAGS.num_epochs,
validation_data=(x_test, y_test),
class_weight=class_weight_dict,
verbose=1,
callbacks=[
checkpointer, tensorboard, loss_history, evaluate_step, clr,
metrics
])
print('-' * 30)
time_str = datetime.datetime.now().isoformat()
print("{}: Done training.".format(time_str))
K.clear_session()
print('-' * 30)
print()