def train(self, dataset=None, initial_epoch=0):
callbacks = get_callbacks(self.model_name, self.tb_path,
self.model_path_new_train,
self.config['lr_dec'], self.config['lr'])
if dataset == None:
dataset = Dataset(self.data_name, self.config_path)
dataset_train, dataset_val = dataset.get_tf_data()
self.model.compile(optimizer=tf.keras.optimizers.Adam(
learning_rate=self.config['lr']),
loss=[marginLoss, 'mse'],
loss_weights=[1., self.config['lmd_gen']],
metrics={'Original_CapsNet': 'accuracy'})
print('-' * 30 + f'{self.data_name} train' + '-' * 30)
history = self.model.fit(dataset_train,
epochs=self.config['epochs'],
validation_data=(dataset_val),
batch_size=self.config['batch_size'],
initial_epoch=initial_epoch,
callbacks=callbacks,
workers=self.config['num_workers'])
self.model.save_weights(
os.path.join(self.config['saved_model_dir'], f"{self.model_name}",
f"{self.model_name}_{self.data_name}.h5"))
return history
def train(self, dataset=None, initial_epoch=0):
callbacks = get_callbacks(self.model_name,
self.tb_path,
self.model_path_new_train,
self.config['lr_dec'],
self.config['lr'],
optimizer=self.optimizer)
if dataset is None:
dataset = Dataset(self.data_name, self.config_path)
dataset_train, dataset_val = dataset.get_tf_data()
if self.optimizer == 'Adam':
self.model.compile(optimizer=tf.keras.optimizers.Adam(
lr=self.config['lr'], momentum=0.9),
loss=[marginLoss, 'mse'],
loss_weights=[1., self.config['lmd_gen']],
metrics={self.model_name: 'accuracy'})
else:
self.model.compile(
optimizer=tf.keras.optimizers.SGD(lr=self.config['lr']),
loss=[marginLoss, 'mse'],
loss_weights=[1., self.config['lmd_gen']],
metrics={self.model_name: 'accuracy'})
# self.model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=self.config['lr']),
# loss=[marginLoss, 'mse'],
# loss_weights=[1., self.config['lmd_gen']],
# metrics={self.model_name: 'accuracy'})
steps = None
print('-' * 30 + f'{self.data_name} train' + '-' * 30)
history = self.model.fit(dataset_train,
epochs=self.config[f'epochs'],
steps_per_epoch=steps,
validation_data=dataset_val,
batch_size=self.config['batch_size'],
initial_epoch=initial_epoch,
callbacks=callbacks,
workers=self.config['num_workers'])
self.model.save_weights(
os.path.join(self.config['saved_model_dir'], f"{self.model_name}",
f"{self.model_name}_{self.data_name}.h5"))
return history
def train(self, dataset=None, initial_epoch=0, resume=False):
callbacks = get_callbacks(self.model_path_new_train,
optimizer=self.optimizer)
if dataset is None:
dataset = Dataset(self.data_name, self.config_path)
dataset_train, dataset_val = dataset.get_tf_data(for_capsule=False)
if self.optimizer == 'SGD':
self.model.compile(optimizer=tf.keras.optimizers.SGD(
lr=self.config['ETC_MODEL_LR'], momentum=0.9),
loss='categorical_crossentropy',
metrics='accuracy')
elif self.optimizer == 'Adam':
self.model.compile(optimizer=tf.keras.optimizers.Adam(
lr=self.config['ETC_MODEL_LR']),
loss='categorical_crossentropy',
metrics='accuracy')
else:
print("optimizer must be select in ['Adam', 'SGD']")
raise ValueError
steps = None
if resume:
self.load_graph_weights()
print('-' * 30 + f'{self.data_name} train' + '-' * 30)
history = self.model.fit(dataset_train,
epochs=self.config[f'ETC_MODEL_EPOCHS'],
steps_per_epoch=steps,
validation_data=dataset_val,
batch_size=self.config['batch_size'],
initial_epoch=initial_epoch,
callbacks=callbacks,
workers=self.config['num_workers'])
self.model.save_weights(
os.path.join(self.config['saved_model_dir'], f"{self.model_name}",
f"{self.model_name}_{self.data_name}.h5"))
return history
def train(self, dataset=None, initial_epoch=0):
callbacks = get_callbacks(self.tb_path, self.model_path_new_train,
self.config['lr_dec'], self.config['lr'])
if dataset == None:
dataset = Dataset(self.model_name, self.config_path)
dataset_train, dataset_val = dataset.get_tf_data()
if self.model_name == 'MULTIMNIST':
self.model.compile(optimizer=tf.keras.optimizers.Adam(
learning_rate=self.config['lr']),
loss=[marginLoss, 'mse', 'mse'],
loss_weights=[
1., self.config['lmd_gen'] / 2,
self.config['lmd_gen'] / 2
],
metrics={'Efficient_CapsNet': multiAccuracy})
steps = 10 * int(
dataset.y_train.shape[0] / self.config['batch_size'])
else:
self.model.compile(optimizer=tf.keras.optimizers.Adam(
learning_rate=self.config['lr']),
loss=[marginLoss, 'mse'],
loss_weights=[1., self.config['lmd_gen']],
metrics={'Efficient_CapsNet': 'accuracy'})
steps = None
print('-' * 30 + f'{self.model_name} train' + '-' * 30)
history = self.model.fit(dataset_train,
epochs=self.config[f'epochs'],
steps_per_epoch=steps,
validation_data=(dataset_val),
batch_size=self.config['batch_size'],
initial_epoch=initial_epoch,
callbacks=callbacks)
return history
def train(self, dataset=None, initial_epoch=0):
if dataset is None:
dataset = Dataset(self.model_name, self.config_path)
dataset_train, dataset_val = dataset.get_tf_data()
self.model.compile(optimizer=tf.keras.optimizers.Adam(
learning_rate=self.config['lr']),
loss=[marginLoss, 'mse', 'mse'],
loss_weights=[
1., self.config['lmd_gen'] / 2,
self.config['lmd_gen'] / 2
],
metrics={'Efficient_CapsNet': 'accuracy'})
steps = 10 * int(dataset.y_train.shape[0] / self.config['batch_size'])
print('-' * 30 + f'{self.model_name} train' + '-' * 30)
history = self.model.fit(dataset_train,
epochs=self.config['epoch'],
steps_per_epoch=steps,
validation_data=(dataset_val),
batch_size=self.config['batch_size'],
initial_epoch=initial_epoch)
return history
