def train():
model = AttentionModel(params_config, human_vocab_size,
machine_vocab_size).model
op = Adam(lr=params_config['learning_rate'],
decay=params_config['decay'],
clipnorm=params_config['clipnorm'])
if os.path.exists('./Model/model.h5'):
print('loading model...')
model.load_weights('./Model/model.h5')
model.compile(optimizer=op,
loss='categorical_crossentropy',
metrics=['accuracy'])
else:
model.compile(optimizer=op,
loss='categorical_crossentropy',
metrics=['accuracy'])
outputs_train = list(Yoh_train.swapaxes(0, 1))
model.fit(Xoh_train,
outputs_train,
epochs=params_config['epochs'],
batch_size=params_config['batch_size'],
validation_split=0.1)
if not os.path.exists('Model'):
os.mkdir('Model')
model.save_weights('./Model/model.h5')
return model
def load_model(path, embed_dim=128, n_customer=20, n_encode_layers=3):
""" Load model weights from hd5 file
https://stackoverflow.com/questions/51806852/cant-save-custom-subclassed-model
"""
small_dataset = generate_data(n_samples=5, n_customer=n_customer)
model_loaded = AttentionModel(embed_dim, n_encode_layers=n_encode_layers)
for data in (small_dataset.batch(5)):
_, _ = model_loaded(data, decode_type='greedy')
model_loaded.load_weights(path)
return model_loaded
def load_model(path, embed_dim=128, n_customer=20, n_encode_layers=3):
""" Load model weights from hd5 file
https://stackoverflow.com/questions/51806852/cant-save-custom-subclassed-model
"""
CAPACITIES = {10: 20., 20: 30., 50: 40., 100: 50.}
data_random = (
tf.random.uniform((2, 2), minval=0, maxval=1),
tf.random.uniform((2, n_customer, 2), minval=0, maxval=1),
tf.cast(
tf.random.uniform(
(2, n_customer), minval=1, maxval=10, dtype=tf.int32),
tf.float32) / tf.cast(CAPACITIES[n_customer], tf.float32))
model_loaded = AttentionModel(embed_dim,
n_encode_layers=n_encode_layers,
decode_type='greedy')
_, _ = model_loaded(data_random)
model_loaded.load_weights(path)
return model_loaded