TestX = np.concatenate(
(TestX, utils.multi_prep(TestXt, target_sig_length)))
TrainY = np.concatenate((TrainY, TrainYt))
TestY = np.concatenate((TestY, TestYt))
del TrainXt, TestXt, TrainYt, TestYt
TrainX = np.expand_dims(TrainX, axis=2)
TestX = np.expand_dims(TestX, axis=2)
TrainY = to_categorical(TrainY, num_classes=len(target_class))
TestY = to_categorical(TestY, num_classes=len(target_class))
toc = time.time()
print('Time for data processing--- ' + str(toc - tic) + ' seconds---')
# ------------------------ 网络生成与训练 ----------------------------
model = net.build_network(config)
model.summary()
model_name = 'myNet.h5'
checkpoint = ModelCheckpoint(filepath=model_name,
monitor='val_categorical_accuracy',
mode='max',
save_best_only='True')
lr_scheduler = LearningRateScheduler(config.lr_schedule)
callback_lists = [checkpoint, lr_scheduler]
model.fit(x=TrainX,
y=TrainY,
batch_size=config.batch_size,
epochs=50,
verbose=1,
validation_data=(TestX, TestY),
def bits_loss(msg, output, message_length):
"""Autoencoder error in number of different bits."""
return reconstruction_loss(msg, output) * message_length
message_length = 16 # in bits
key_length = message_length # in bits
batch = 512 # Number of messages to train on at once
adv_iter = 100 # Adversarial iterations
max_iter = 20 # Individual agent iterations
learning_rate = 0.0008
if __name__ == "__main__":
msg, key = build_input_layers(message_length, key_length)
alice_output, bob_output, eve_output = build_network(msg, key)
eve_loss = reconstruction_loss(msg, eve_output)
bob_reconst_loss = reconstruction_loss(msg, bob_output)
bob_loss = bob_reconst_loss + (0.5 - eve_loss)**2
eve_bit_loss = bits_loss(msg, eve_output, message_length)
bob_bit_loss = bits_loss(msg, bob_output, message_length)
AB_vars = (tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, "alice") +
tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, "bob"))
E_vars = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, "eve")
trainAB = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(
bob_loss, var_list=AB_vars)