output_list.append(model_name + "\n") print(model_name) output[model_name][str(reliability_setting)][ iteration - 1] = calculateExpectedAccuracy( model, no_information_flow_map, reliability_setting, output_list, training_labels=training_labels, test_data=test_data, test_labels=test_labels) # runs all 3 failure configurations for all 3 models if __name__ == "__main__": accuracy = accuracy("Health") calculateExpectedAccuracy = accuracy.calculateExpectedAccuracy use_GCP = False training_data, val_data, test_data, training_labels, val_labels, test_labels = init_data( use_GCP) ResiliNet_no_information_flow_map = make_no_information_flow_map( "Health", default_skip_hyperconnection_config) deepFogGuard_no_information_flow_map = make_no_information_flow_map( "Health", default_skip_hyperconnection_config) Vanilla_no_information_flow_map = make_no_information_flow_map("Health") num_iterations, num_vars, num_classes, reliability_settings, num_train_epochs, hidden_units, batch_size = init_common_experiment_params( training_data) load_for_inference = False
input_shape=input_shape, alpha=alpha, reliability_setting=reliability_setting, skip_hyperconnection_config=skip_hyperconnection_configuration, strides=strides, num_gpus=num_gpus) get_model_weights_CNN_cifar( model, parallel_model, model_name, load_for_inference, model_file, training_data, training_labels, val_data, val_labels, train_datagen, batch_size, epochs, progress_verbose, checkpoint_verbose, train_steps_per_epoch, val_steps_per_epoch, num_gpus) return model if __name__ == "__main__": accuracy = accuracy("CIFAR") calculateExpectedAccuracy = accuracy.calculateExpectedAccuracy training_data, test_data, training_labels, test_labels, val_data, val_labels = init_data( ) num_iterations, classes, reliability_settings, train_datagen, batch_size, epochs, progress_verbose, checkpoint_verbose, use_GCP, alpha, input_shape, strides, num_gpus = init_common_experiment_params( ) skip_hyperconnection_configurations = [ # [e1,IoT] [0, 0], [1, 0], [0, 1], [1, 1], ] model_name = "ResiliNet Hyperconnection Weight Sensitivity" default_reliability_setting = [1.0, 1.0, 1.0]
hidden_units, reliability_setting=reliability_setting, hyperconnection_weights_scheme=weight_scheme) model_file = "models/" + str(iteration) + "_" + str( reliability_setting) + "_" + str( weight_scheme) + 'camera_hyperconnection_ResiliNet.h5' get_model_weights_MLP_camera(model, model_name, load_for_inference, model_file, training_data, training_labels, val_data, val_labels, num_train_epochs, batch_size, verbose) return model # runs all 3 failure configurations for all 3 models if __name__ == "__main__": accuracy = accuracy("Camera") calculateExpectedAccuracy = accuracy.calculateExpectedAccuracy use_GCP = False training_data, val_data, test_data, training_labels, val_labels, test_labels = init_data( use_GCP) reliability_settings, input_shape, num_classes, hidden_units, batch_size, num_train_epochs, num_iterations = init_common_experiment_params( ) no_information_flow_map = make_no_information_flow_map( "Camera", default_skip_hyperconnection_config) load_for_inference = False # file name with the experiments accuracy output output_name = "results/camera_hyperconnection_weight.txt" verbose = 2 model_name = "ResiliNet Hyperconnection Weight"
num_test_examples): output_list.append(model_name + "\n") print(model_name) output[model_name][str(reliability_setting)][ iteration - 1] = calculateExpectedAccuracy( model, no_information_flow_map, reliability_setting, output_list, test_generator=test_generator, num_test_examples=num_test_examples) # runs all 3 failure configurations for all 3 models if __name__ == "__main__": accuracy = accuracy("Imagenet") calculateExpectedAccuracy = accuracy.calculateExpectedAccuracy use_GCP = False num_iterations, num_train_examples, num_test_examples, reliability_settings, input_shape, num_classes, alpha, epochs, num_gpus, strides, num_workers = init_common_experiment_params( ) train_generator, test_generator = init_data(use_GCP, num_gpus) ResiliNet_no_information_flow_map = make_no_information_flow_map( "CIFAR/Imagenet", default_skip_hyperconnection_config) deepFogGuard_no_information_flow_map = make_no_information_flow_map( "CIFAR/Imagenet", default_skip_hyperconnection_config) Vanilla_no_information_flow_map = make_no_information_flow_map( "CIFAR/Imagenet") load_for_inference = False continue_training = False # loads a pre-trained model and improves it with more training