def run(model, train_set=[], test_sets=[]):
dataController = DataController(batch_size=batch_size, data_list=train_set)
logging.info("Training on: {}".format(train_set))
model.train(dataController)
for test_set in test_sets:
dataController = DataController(batch_size=batch_size, data_list=test_set)
model.load()
if validation_mode == 'OpenMax':
model.calc_mean_and_dist(dataController, 4)
logging.info("Validate on: {}".format(test_set))
model.validate(dataController, 4, mode=validation_mode)
def prepare_data(labels):
datas = {}
autoencoder = keras.models.load_model('model')
layer_name = 'enc'
encoded_layer = Model(inputs=autoencoder.input,
outputs=autoencoder.get_layer(layer_name).output)
for label in labels:
print("Getting encoded data for", label)
datas[label] = []
dataController = DataController(batch_size=batch_size,
data_list=normal_labels)
while 1:
data = dataController.generate('full')
if data is False:
break
x = data["x"]
enc_out = encoded_layer.predict(x)
datas[label].extend(enc_out)
datas[label] = np.array(datas[label])
with open('encoded_datas.pkl', 'wb') as f:
pickle.dump(datas, f, pickle.HIGHEST_PROTOCOL)
return datas
