def save_classifier(self, debug, filename=None, filepath=None):
"""
Saves projections classifiers.
"""
if filename is None or filepath is None:
filename, filepath = directories._get_model_savedir(
model_name=self.model_name,
dataset_name=self.dataset_name,
size_proj=self.size_proj,
projection_mode=self.projection_mode,
epochs=self.epochs,
centroid_translation=self.centroid_translation,
debug=debug)
if self.trained:
for i, seed in enumerate(self.random_seeds):
# for seed, proj_classifier in enumerate(self.classifiers):
proj_classifier = self.classifiers[i]
proj_classifier.save_classifier(filepath=filepath,
filename=filename + "_" +
str(seed),
debug=debug)
if self.centroid_translation:
save_to_pickle(data=self.translation_vector,
filepath=filepath,
filename="training_data_centroid.pkl")
else:
raise ValueError("Train the model first.")
def load_classifier(self, debug, filename=None, filepath=None):
if filename is None or filepath is None:
filename, filepath = directories._get_model_savedir(model_name=self.model_name, dataset_name=self.dataset_name,
epochs=self.epochs, debug=debug)
return super(BaselineConvnet, self).load_classifier(filepath=filepath, filename=filename)
def save_robust_classifier(self, robust_classifier, debug, attack_method, attack_library, filename=None, filepath=None):
if filename is None or filepath is None:
filename, filepath = directories._get_model_savedir(model_name=self.model_name, dataset_name=self.dataset_name,
epochs=self.epochs, debug=debug, robust=True,
attack_method=attack_method, attack_library=attack_library)
robust_classifier.save_classifier(filepath=filepath, filename=filename, debug=debug)
def load_robust_classifier(self, debug, attack_method, attack_library, filename=None, filepath=None):
if filename is None or filepath is None:
filename, filepath = directories._get_model_savedir(model_name=self.model_name, dataset_name=self.dataset_name,
epochs=self.epochs, debug=debug, robust=True,
attack_method=attack_method, attack_library=attack_library)
robust_classifier = BaselineConvnet(input_shape=self.input_shape, num_classes=self.num_classes, epochs=self.epochs,
data_format=self.data_format, dataset_name=self.dataset_name)
robust_classifier.load_classifier(filepath=filepath, filename=filename, debug=debug)
return robust_classifier
def load_classifier(self, debug, filename=None, filepath=None):
"""
Loads a pre-trained classifier.
"""
if filename is None or filepath is None:
filename, filepath = directories._get_model_savedir(
model_name=self.model_name,
dataset_name=self.dataset_name,
size_proj=self.size_proj,
projection_mode=self.projection_mode,
epochs=self.epochs,
centroid_translation=self.centroid_translation,
debug=debug)
start_time = time.time()
self.trained = True
classifiers = []
for seed in self.random_seeds:
proj_classifier = BaselineConvnet(input_shape=self.input_shape,
num_classes=self.num_classes,
epochs=self.epochs,
data_format=self.data_format,
dataset_name=self.dataset_name)
classifiers.append(
proj_classifier.load_classifier(filepath=filepath,
filename=filename + "_" +
str(seed),
debug=debug))
if self.centroid_translation:
self.translation_vector = load_from_pickle(
path=filepath + "training_data_centroid.pkl")
else:
self.translation_vector = None
print("\nLoading time: --- %s seconds ---" %
(time.time() - start_time))
self.classifiers = classifiers
return classifiers