def save_png(self, save_dir, epoch):
if torch.cuda.is_available():
noise = self.noise.cuda()
with torch.no_grad():
gen_data = self.G(noise)
gen_data = gen_data.cpu().detach().numpy()
img_bhwc = np.transpose(gen_data, (0, 2, 3, 1))
img_tile = plotting.img_tile(img_bhwc,
aspect_ratio=1.0,
border_color=1.0,
stretch=True)
img = plotting.plot_img(img_tile,
title='CIFAR10 samples ' + str(epoch) +
" epochs")
plotting.plt.savefig(save_dir + "cifar_sample_feature_match_" +
str(epoch) + ".png")
parser.add_argument('--data_dir', type=str, default='~/temp/data')
parser.add_argument('--plot_title', type=str, default=None)
args = parser.parse_args()
print(args)
data_dir = args.data_dir
trainx, trainy = cifar10_data.load(data_dir)
ids = [[] for i in range(10)]
for i in range(trainx.shape[0]):
if len(ids[trainy[i]]) < 10:
ids[trainy[i]].append(i)
if np.alltrue(np.asarray([len(_ids) >= 10 for _ids in ids])):
break
images = np.zeros((10 * 10, 32, 32, 3), dtype='uint8')
for i in range(len(ids)):
for j in range(len(ids[i])):
images[10 * j + i] = trainx[ids[i][j]].transpose([1, 2, 0])
print(ids)
img_tile = plotting.img_tile(images,
aspect_ratio=1.0,
border_color=1.0,
stretch=True)
img = plotting.plot_img(
img_tile, title=args.plot_title if args.plot_title != 'None' else None)
plotting.plt.savefig(args.save_dir + '/cifar10_orig_images.png')
plotting.plt.close('all')
else:
c += 1
if c == pat: break
# log progress to console
print(
"Iteration %d, time = %ds, train bits_per_dim = %.4f, test bits_per_dim = %.4f"
% (epoch, time.time() - begin, train_loss_gen, test_loss_gen))
sys.stdout.flush()
if epoch % args.save_interval == -1:
print("Generating samples...")
# generate samples from the model
sampled_x = sample_from_model(sess)
img_tile = plotting.img_tile(sampled_x,
aspect_ratio=1.0,
border_color=1.0,
stretch=True)
img = plotting.plot_img(img_tile, title=args.data_set + ' samples')
plotting.plt.savefig(
os.path.join(args.save_dir,
'%s_sample%d.png' % (args.data_set, epoch)))
plotting.plt.close('all')
# save params
saver.save(sess,
args.save_dir + '/params_' + args.data_set + '.ckpt')
np.savez(args.save_dir + '/test_bpd_' + args.data_set + '.npz',
test_bpd=np.array(test_bpd))
print("online: ", first)
print("offline: ", testl)
loss_unl += lu
train_err += te
train_err2 +=te2
e = train_batch_gen(trainx_unl2[t*args.batch_size:(t+1)*args.batch_size],lr) # disc and gen for unlabeled data are different
gen_loss += float(e)
#clamp_D_fn()
loss_lab /= nr_batches_train
loss_unl /= nr_batches_train
train_err /= nr_batches_train
train_err2 /= nr_batches_train
# test
test_err = 0.
for t in range(nr_batches_test):
test_err += test_batch(testx[t*args.batch_size:(t+1)*args.batch_size],testy[t*args.batch_size:(t+1)*args.batch_size])
test_err /= nr_batches_test
# report
print("Iteration %d, time = %ds, loss_lab = %.4f, loss_unl = %.4f, train err = %.4f, train err2 = %.4f,gen loss = %.4f,test err = %.4f" % (epoch, time.time()-begin, loss_lab, loss_unl, train_err,train_err2,gen_loss,test_err))
sys.stdout.flush()
# generate samples from the model
sample_x = samplefun()
img_bhwc = np.transpose(sample_x[:100,], (0, 2, 3, 1))
img_tile = plotting.img_tile(img_bhwc, aspect_ratio=1.0, border_color=1.0, stretch=True)
img = plotting.plot_img(img_tile, title='CIFAR10 samples')
plotting.plt.savefig("cifar_sample_feature_match_50_pai.png")
# save params
np.savez('disc_params.npz', *[p.get_value() for p in disc_params])
np.savez('gen_params.npz', *[p.get_value() for p in gen_params])
test_list.append(test_err)
# report
print(
"Epoch %d, time = %ds, loss_lab = %.3f, learning_rate = %.6f,train err= %.4f, test err = %.4f"
% (epoch, time.time() - begin, loss_lab, learning_rate_var.get_value(),
train_err, test_err))
sys.stdout.flush()
# generate samples from the model
sample_y = np.int32(
np.random.choice(num_classes, size=(args.batch_size), replace=False))
sample_x = samplefun(sample_y)
img_bhwc = np.transpose(sample_x[:100, ], (0, 2, 3, 1))
img_tile = plotting.img_tile(img_bhwc,
aspect_ratio=1.0,
border_color=1.0,
stretch=True)
img = plotting.plot_img(img_tile, title='CIFAR-100 samples')
plotting.plt.savefig(args.results_dir + args.dataset +
'/cifar100_sample.png')
if epoch % 20 == 0:
NNdiff = np.sum(np.sum(np.sum(np.square(
np.expand_dims(sample_x, axis=1) - np.expand_dims(trainx, axis=0)),
axis=2),
axis=2),
axis=2)
NN = trainx[np.argmin(NNdiff, axis=1)]
NN = np.transpose(NN[:100], (0, 2, 3, 1))
NN_tile = plotting.img_tile(NN,
aspect_ratio=1.0,
border_color=1.0,
