def load_valid_dataset(dataset_type): global valid_dataset if dataset_type == 'cifar-10': valid_dataset_filename = '..' + os.sep + 'data' + os.sep + 'cifar-10.pickle' (train_dataset, train_labels), \ (valid_dataset, valid_labels), \ (test_dataset, test_labels) = load_data.reformat_data_cifar10(valid_dataset_filename) del train_dataset, train_labels, test_dataset, test_labels elif dataset_type == 'imagenet-100': valid_dataset_fname = 'imagenet_small' + os.sep + 'imagenet_small_valid_dataset' valid_label_fname = 'imagenet_small' + os.sep + 'imagenet_small_valid_labels' fp1 = np.memmap(valid_dataset_fname, dtype=np.float32, mode='r', offset=np.dtype('float32').itemsize * 0, shape=(valid_size, image_size, image_size, num_channels)) fp2 = np.memmap(valid_label_fname, dtype=np.int32, mode='r', offset=np.dtype('int32').itemsize * 0, shape=(valid_size, 1)) v_dataset = fp1[:, :, :, :] v_labels = fp2[:] v_dataset, v_labels = load_data.reformat_data_imagenet_with_memmap_array( v_dataset, v_labels, silent=True) del v_labels valid_dataset = v_dataset
beta = 1e-3 try: opts,args = getopt.getopt( sys.argv[1:],"",['data=',"log_suffix="]) except getopt.GetoptError as err: print('<filename>.py --data= --log_suffix=') if len(opts)!=0: for opt,arg in opts: if opt == '--data': data_filename = arg if opt == '--log_suffix': log_suffix = arg if dataset_type=='cifar-10': (full_train_dataset,full_train_labels),(valid_dataset,valid_labels),(test_dataset,test_labels)=load_data.reformat_data_cifar10(data_filename) graph = tf.Graph() # Value logger will log info used to calculate policies test_logger = logging.getLogger('test_logger_'+log_suffix) test_logger.setLevel(logging.INFO) fileHandler = logging.FileHandler('test_logger_'+log_suffix, mode='w') fileHandler.setFormatter(logging.Formatter('%(message)s')) test_logger.addHandler(fileHandler) test_accuracies = [] with tf.Session(graph=graph) as session: #tf.global_variables_initializer().run() # Input data.
__author__ = 'Thushan Ganegedara' import load_data from scipy.misc import imsave import numpy as np if __name__=='__main__': load_data.load_and_save_data_cifar10(filename='cifar-10-white.pickle',zca_whiten=True,return_original=True,separate_rgb=False) (tr_white_dataset,tr_labels),(v_white_dataset,v_labels),(ts_white_dataset,ts_labels) = load_data.reformat_data_cifar10(filename='cifar-10-white.pickle') #(tr_dataset,tr_labels),(v_dataset,v_labels),(ts_dataset,ts_labels) = load_data.reformat_data_cifar10(filename='cifar-10.pickle') for i in range(10): rand_idx = np.random.randint(0,5000) #imsave('test_img.png', tr_dataset[rand_idx,:,:,:]) imsave('test_img_whitened_'+str(i)+'.png', tr_white_dataset[rand_idx,:,:,:])