klds)
print("epoch {0} --------------------- Time {1:.2f}s".format(
epoch,
time.time() - tt))
print(
"train loss:{0:.3f}, train mse:{1:.3f}, train kld:{2:.3f}".format(
train_loss, train_mse, train_kld))
print("test loss:{0:.3f}, test mse:{1:.3f}, test kld:{2:.3f}".format(
test_loss, test_mse, test_kld))
if epoch % FLAGS.save_interval == 0:
saver.save(sess, FLAGS.save_dir + '/params_' + 'celeba' + '.ckpt')
data = next(test_data)
feed_dict = make_feed_dict(data, test_mgen)
sample_x = sess.run(x_hats, feed_dict=feed_dict)
sample_x = np.concatenate(sample_x, axis=0)
test_data.reset()
img_tile = plotting.img_tile(sample_x[:25],
aspect_ratio=1.0,
border_color=1.0,
stretch=True)
img = plotting.plot_img(img_tile,
title=FLAGS.data_set + ' samples')
plotting.plt.savefig(
os.path.join(FLAGS.save_dir,
'%s_vae_sample%d.png' % (FLAGS.data_set, epoch)))
sess.run(initializer)
feed_dict = make_feed_dict(
train_data.next(args.init_batch_size), init=True
) # manually retrieve exactly init_batch_size examples
sess.run(init_pass, feed_dict)
# generate samples from the model
sample_x = []
for i in range(args.num_samples):
sample_x.append(sample_from_model(sess))
sample_x = np.concatenate(sample_x, axis=0)
img_tile = plotting.img_tile(sample_x[:100],
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')
np.savez(
os.path.join(args.save_dir,
'%s_sample%d.npz' % (args.data_set, epoch)),
sample_x)
print('starting training')
# train for one epoch
train_losses = []
cnt = 0
for d in train_data:
# compute likelihood over test data
test_losses = []
for d in test_data:
feed_dict = make_feed_dict(d)
l = sess.run(bits_per_dim_test, feed_dict)
test_losses.append(l)
test_loss_gen = np.mean(test_losses)
test_bpd.append(test_loss_gen)
# 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 == 0:
# generate samples from the model
sample_x = []
for i in range(args.num_samples):
sample_x.append(sample_from_model(sess))
sample_x = np.concatenate(sample_x,axis=0)
img_tile = plotting.img_tile(sample_x[:100], aspect_ratio=1.0, border_color=1.0, stretch=True)
img = plotting.plot_img(img_tile, title=dset_dump_name + ' samples')
plotting.plt.savefig(os.path.join(args.save_dir,'%s_sample%d.png' % (dset_dump_name, epoch)))
plotting.plt.close('all')
np.savez(os.path.join(args.save_dir,'%s_sample%d.npz' % (dset_dump_name, epoch)), sample_x)
# save params
saver.save(sess, args.save_dir + '/params_' + dset_dump_name + '.ckpt', global_step=epoch)
np.savez(args.save_dir + '/test_bpd_' + dset_dump_name + '.npz', test_bpd=np.array(test_bpd))