print(name) print(max_iter) #sys.exit() CV_SPLITS = 10 # LOAD DATA #dsBunch = ds.load('iris') #data = train_test_split(dsBunch.data, dsBunch.target, test_size=0.25, random_state=1) # dsBunch = ds.load_mnist_back() # dsTest = ds.load_mnist_back_test() # data = (dsBunch.data, dsTest.data, dsBunch.target, dsTest.target) X_train, y_train = ds.load_mnist('data') X_test, y_test = ds.load_mnist('data', kind='t10k') data = (X_train[:10000], X_test, y_train[:10000], y_test) # #print('DATA:') #n_features = dsBunch.data.shape[1] #shp = dsBunch.data.shape #print(pd.DataFrame(dsTest.data).head) #print() #print(pd.DataFrame(dsTest.target).head) #print('n_features: {}\nshape: {}\n'.format(n_features, shp)) # DEFINE PARAM GRIDS # d_features = n_features//2 # hls = [(d_features,)*3, (n_features,)*3, (d_features,)*2, (n_features,)*2, (d_features,), (n_features,),] hls = [
############################# ### Preliminaries ############################# # Retrieve the arguments from the command-line args = parseArgs() # Fix the seed for the random generator np.random.seed(seed=0) ############################# ### Dataset Handling ############################# ### Load the dataset train_set, valid_set, test_set = dataset_loader.load_mnist() ### Define the dataset variables n_training = train_set[0].shape[0] n_feature = train_set[0].shape[1] n_label = np.max(train_set[1])+1 ############################# ### Neural Network parameters ############################# ### Activation function act_func_name = args.act_func ### Network Architecture nn_arch = np.array([n_feature] + args.arch + [n_label])
*convblock(128, 256, 4, 2, 1), nn.Conv2d(256, 1, 4, 1, 0, bias=False), # FC with Conv. nn.Sigmoid() ) def forward(self, img): prob = self.model(img) return prob assert (opt.dataset == 'cifar10' or opt.dataset == 'mnist'), 'Unknown dataset! Only cifar10 and mnist are supported.' if opt.dataset == 'cifar10': batch_iterator = DataLoader(load_cifar10(opt.img_size), shuffle=True, batch_size=opt.batch_size) # List, NCHW format. elif opt.dataset == 'mnist': batch_iterator = DataLoader(load_mnist(opt.img_size), shuffle=True, batch_size=opt.batch_size) # List, NCHW format. # Save a batch of real images for reference. os.makedirs('./out', exist_ok=True) save_image(next(iter(batch_iterator))[0][:25, ...], './out/real_samples.png', nrow=5, normalize=True) cuda = torch.cuda.is_available() Tensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor gan_loss = nn.BCELoss() generator = Generator() discriminator = Discriminator() optimizer_D = optim.RMSprop(discriminator.parameters(), lr=opt.lr) optimizer_G = optim.RMSprop(generator.parameters(), lr=opt.lr)