def main(_):
np.random.seed(0)
tf.set_random_seed(0)
pp.pprint(flags.FLAGS.__flags)
if FLAGS.input_width is None:
FLAGS.input_width = FLAGS.input_height
if not os.path.exists(FLAGS.checkpoint_dir):
os.makedirs(FLAGS.checkpoint_dir)
if not os.path.exists(FLAGS.sample_dir):
os.makedirs(FLAGS.sample_dir)
run_config = tf.ConfigProto()
run_config.gpu_options.allow_growth = True
run_config.allow_soft_placement = True
sess = None
# main.py - -dataset
# mnist - -input_height = 28 - -output_height = 28 - -train
# - -dataset
# mnist - -input_height = 28 - -output_height = 28 - -train
flags2 = tf.app.flags
#flags2.DEFINE_integer("epoch2", 25, "Epoch to train [25]")
flags2.DEFINE_integer("epoch2", 251, "Epoch to train [25]")
flags2.DEFINE_float("learning_rate2", 0.0002,
"Learning rate of for adam [0.0002]")
flags2.DEFINE_float("beta12", 0.5, "Momentum term of adam [0.5]")
flags2.DEFINE_float("train_size2", np.inf,
"The size of train images [np.inf]")
#flags2.DEFINE_integer("batch_size2", 1024, "The size of batch images [64]")
flags2.DEFINE_integer("batch_size2", 512, "The size of batch images [64]")
#flags2.DEFINE_integer("batch_size2", 256, "The size of batch images [64]")
flags2.DEFINE_integer(
"input_height2", 28,
"The size of image to use (will be center cropped). [108]")
flags2.DEFINE_integer(
"input_width2", None,
"The size of image to use (will be center cropped). If None, same value as input_height [None]"
)
flags2.DEFINE_integer("output_height2", 28,
"The size of the output images to produce [64]")
flags2.DEFINE_integer(
"output_width2", None,
"The size of the output images to produce. If None, same value as output_height [None]"
)
flags2.DEFINE_string("dataset2", "mnist",
"The name of dataset [celebA, mnist, lsun]")
flags2.DEFINE_string("input_fname_pattern2", "*.jpg",
"Glob pattern of filename of input images [*]")
flags2.DEFINE_string("data_dir2", "./data",
"path to datasets [e.g. $HOME/data]")
flags2.DEFINE_string("out_dir2", "./out",
"Root directory for outputs [e.g. $HOME/out]")
flags2.DEFINE_string(
"out_name2", "",
"Folder (under out_root_dir) for all outputs. Generated automatically if left blank []"
)
flags2.DEFINE_string(
"checkpoint_dir2", "checkpoint",
"Folder (under out_root_dir/out_name) to save checkpoints [checkpoint]"
)
flags2.DEFINE_string(
"sample_dir2", "samples",
"Folder (under out_root_dir/out_name) to save samples [samples]")
flags2.DEFINE_boolean("train2", True,
"True for training, False for testing [False]")
flags2.DEFINE_boolean("crop2", False,
"True for training, False for testing [False]")
flags2.DEFINE_boolean("visualize2", False,
"True for visualizing, False for nothing [False]")
flags2.DEFINE_boolean("export2", False,
"True for exporting with new batch size")
flags2.DEFINE_boolean("freeze2", False,
"True for exporting with new batch size")
flags2.DEFINE_integer("max_to_keep2", 1,
"maximum number of checkpoints to keep")
flags2.DEFINE_integer("sample_freq2", 200,
"sample every this many iterations")
flags2.DEFINE_integer("ckpt_freq2", 200,
"save checkpoint every this many iterations")
flags2.DEFINE_integer("z_dim2", 100, "dimensions of z")
flags2.DEFINE_string("z_dist2", "uniform_signed",
"'normal01' or 'uniform_unsigned' or uniform_signed")
flags2.DEFINE_boolean("G_img_sum2", False,
"Save generator image summaries in log")
# flags.DEFINE_integer("generate_test_images", 100, "Number of images to generate during test. [100]")
FLAGS2 = flags2.FLAGS
pp.pprint(flags2.FLAGS.__flags)
# expand user name and environment variables
FLAGS2.data_dir2 = expand_path(FLAGS2.data_dir2)
FLAGS2.out_dir2 = expand_path(FLAGS2.out_dir2)
FLAGS2.out_name2 = expand_path(FLAGS2.out_name2)
FLAGS2.checkpoint_dir2 = expand_path(FLAGS2.checkpoint_dir2)
FLAGS2.sample_dir2 = expand_path(FLAGS2.sample_dir2)
if FLAGS2.output_height2 is None:
FLAGS2.output_height2 = FLAGS2.input_height2
if FLAGS2.input_width2 is None: FLAGS2.input_width2 = FLAGS2.input_height2
if FLAGS2.output_width2 is None:
FLAGS2.output_width2 = FLAGS2.output_height2
# output folders
if FLAGS2.out_name2 == "":
FLAGS2.out_name2 = '{} - {} - {}'.format(
timestamp(),
FLAGS2.data_dir2.split('/')[-1],
FLAGS2.dataset2) # penultimate folder of path
if FLAGS2.train2:
FLAGS2.out_name2 += ' - x{}.z{}.{}.y{}.b{}'.format(
FLAGS2.input_width2, FLAGS2.z_dim2, FLAGS2.z_dist2,
FLAGS2.output_width2, FLAGS2.batch_size2)
FLAGS2.out_dir2 = os.path.join(FLAGS2.out_dir2, FLAGS2.out_name2)
FLAGS2.checkpoint_dir2 = os.path.join(FLAGS2.out_dir2,
FLAGS2.checkpoint_dir2)
FLAGS2.sample_dir2 = os.path.join(FLAGS2.out_dir2, FLAGS2.sample_dir2)
if not os.path.exists(FLAGS2.checkpoint_dir2):
os.makedirs(FLAGS2.checkpoint_dir2)
if not os.path.exists(FLAGS2.sample_dir2): os.makedirs(FLAGS2.sample_dir2)
with open(os.path.join(FLAGS2.out_dir2, 'FLAGS.json'), 'w') as f:
flags_dict = {k: FLAGS2[k].value for k in FLAGS2}
json.dump(flags_dict, f, indent=4, sort_keys=True, ensure_ascii=False)
# gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
run_config = tf.ConfigProto()
run_config.gpu_options.allow_growth = True
with tf.Session(config=run_config) as sess:
dcgan = DCGAN(sess,
input_width=FLAGS.input_width,
input_height=FLAGS.input_height,
batch_size=FLAGS.batch_size,
sample_num=FLAGS.batch_size,
c_dim=FLAGS.c_dim,
z_dim=FLAGS.c_dim * FLAGS.input_height *
FLAGS.input_width,
dataset_name=FLAGS.dataset,
checkpoint_dir=FLAGS.checkpoint_dir,
f_div=FLAGS.f_div,
prior=FLAGS.prior,
lr_decay=FLAGS.lr_decay,
min_lr=FLAGS.min_lr,
model_type=FLAGS.model_type,
log_dir=FLAGS.log_dir,
alpha=FLAGS.alpha,
batch_norm_adaptive=FLAGS.batch_norm_adaptive,
init_type=FLAGS.init_type,
reg=FLAGS.reg,
n_critic=FLAGS.n_critic,
hidden_layers=FLAGS.hidden_layers,
no_of_layers=FLAGS.no_of_layers,
like_reg=FLAGS.like_reg,
df_dim=FLAGS.df_dim)
dcdcdcgan = DCDCDCGAN(sess,
dcDcgan=dcgan,
input_width=FLAGS2.input_width2,
input_height=FLAGS2.input_height2,
output_width=FLAGS2.output_width2,
output_height=FLAGS2.output_height2,
batch_size=FLAGS2.batch_size2,
sample_num=FLAGS2.batch_size2,
y_dim=10,
z_dim=FLAGS2.z_dim2,
dataset_name=FLAGS2.dataset2,
input_fname_pattern=FLAGS2.input_fname_pattern2,
crop=FLAGS2.crop2,
checkpoint_dir=FLAGS2.checkpoint_dir2,
sample_dir=FLAGS2.sample_dir2,
data_dir=FLAGS2.data_dir2,
out_dir=FLAGS2.out_dir2,
max_to_keep=FLAGS2.max_to_keep2)
'''
dcdcdcgan = DCDCDCGAN(
sess,
input_width=FLAGS.input_width,
input_height=FLAGS.input_height,
batch_size=FLAGS.batch_size,
sample_num=FLAGS.batch_size,
c_dim=FLAGS.c_dim,
z_dim=FLAGS.c_dim * FLAGS.input_height * FLAGS.input_width,
dataset_name=FLAGS.dataset,
checkpoint_dir=FLAGS.checkpoint_dir,
f_div=FLAGS.f_div,
prior=FLAGS.prior,
lr_decay=FLAGS.lr_decay,
min_lr=FLAGS.min_lr,
model_type=FLAGS.model_type,
log_dir=FLAGS.log_dir,
alpha=FLAGS.alpha,
batch_norm_adaptive=FLAGS.batch_norm_adaptive,
init_type=FLAGS.init_type,
reg=FLAGS.reg,
n_critic=FLAGS.n_critic,
hidden_layers=FLAGS.hidden_layers,
no_of_layers=FLAGS.no_of_layers,
like_reg=FLAGS.like_reg,
df_dim=FLAGS.df_dim)
'''
#dcgan.train(FLAGS)
dcgan.train(FLAGS)
#print('asdfasf OK asdfasdf')
#print('asdfasf OK asdfasdf')
#print('asdfasf OK asdfasdf')
#print('asdfasf OK asdfasdf')
#print('')
#dcdcdcgan.train(FLAGS2)
#dcdcdcgan.train(FLAGS2)
#print('asdfasdfdfasf OK OK OK asdfsdfsaasdf')
#print('asdfasdfdfasf OK OK OK asdfsdfsaasdf')
#print('asdfasdfdfasf OK OK OK asdfsdfsaasdf')
#print('asdfasdfdfasf OK OK OK asdfsdfsaasdf')
#print('asdfasdasdfasfdfasf This is OK. asdadsfasfsdfsaasdf')
#print('asdfasdasdfasfdfasf This is OK. asdadsfasfsdfsaasdf')
#print('asdfasdasdfasfdfasf This is OK. asdadsfasfsdfsaasdf')
#print('asdfasdasdfasfdfasf This is OK. asdadsfasfsdfsaasdf')
print('')
#test_nlli = self.evaluate_neg_loglikelihood(test_data, config)
#test_nlli = self.evaluate_neg_loglikelihood(test_data, config)
#test_nlli = self.evaluate_neg_loglikelihood(test_data, config)
import dataset_loaders.mnist_loader as mnist_data
#data_X, val_data, test_data, train_dist = mnist_data.load_mnist()
train_data, val_data, test_data, train_dist = mnist_data.load_mnist()
#data_X, val_data, test_data, train_dist = mnist_data.load_mnist()
test_data = np.reshape(test_data, (-1, dcgan.image_size))
#data_X, val_data, test_data, train_dist = mnist_data.load_mnist()
#test_nlli = dcgan.evaluate_neg_loglikelihood(test_data, FLAGS)
#val_nlli = self.evaluate_neg_loglikelihood(val_data, config)
#val_nlli=self.evaluate_neg_loglikelihood(val_data, config)
#val_nlli=self.evaluate_neg_loglikelihood(val_data, config)
#val_nlli = self.evaluate_neg_loglikelihood(val_data, config)
#val_nlli = self.evaluate_neg_loglikelihood(val_data, config)
#print('')
#print('asdfaassdfassdasdfasfdfasf OK. This is OK. OK. asdadsfsasfssafasddfsaasdf')
#print('asdfaassdfassdasdfasfdfasf OK. This is OK. OK. asdadsfsasfssafasddfsaasdf')
#print('asdfaassdfassdasdfasfdfasf OK. This is OK. OK. asdadsfsasfssafasddfsaasdf')
#print('asdfaassdfassdasdfasfdfasf OK. This is OK. OK. asdadsfsasfssafasddfsaasdf')
#print(test_nlli)
#print(test_nlli)
#print('')
#print(test_nlli)
val_data = np.reshape(val_data, (-1, dcgan.image_size))
#val_nlli = dcgan.evaluate_neg_loglikelihood(val_data, FLAGS)
train_data = np.reshape(train_data, (-1, dcgan.image_size))
#train_nlli = dcgan.evaluate_neg_loglikelihood(train_data, FLAGS)
#print(val_nlli)
#print(train_nlli)
#print(val_nlli)
#print(val_nlli)
#print(train_nlli)
#print(train_nlli)
#print(train_nlli)
print('')
# test_data is (10000, 784)
# the size is now (10000, 784)
#nlli_test = sess.run([dcgan.log_likelihood],
# feed_dict={dcgan.log_like_batch: test_data})
#nlli_test = sess.run([dcgan.log_likelihood],
# feed_dict={dcgan.log_like_batch: test_data[0,:].repeat(64, 1)})
# torch.cat([input]*100)
# use: torch.cat([input]*100)
# np.tile(a,(3,1))
# use: np.tile(a,(3,1))
# test_data[0, :]
# use: test_data[0, :]
nlli_test = sess.run([dcgan.log_likelihood],
feed_dict={
dcgan.log_like_batch:
np.tile(test_data[0, :],
(FLAGS.batch_size, 1))
})
nlli_val = sess.run([dcgan.log_likelihood],
feed_dict={
dcgan.log_like_batch:
np.tile(val_data[0, :], (FLAGS.batch_size, 1))
})
nlli_train = sess.run([dcgan.log_likelihood],
feed_dict={
dcgan.log_like_batch:
np.tile(train_data[0, :],
(FLAGS.batch_size, 1))
})
#train_nlli = dcgan.evaluate_neg_loglikelihood(train_data, FLAGS)
#train_nlli = dcgan.evaluate_neg_loglikelihood(train_data, FLAGS)
#train_nlli = dcgan.evaluate_neg_loglikelihood(train_data, FLAGS)
nlli_test = np.squeeze(nlli_test)
nlli_val = np.squeeze(nlli_val)
nlli_train = np.squeeze(nlli_train)
print(nlli_test)
print(nlli_val)
print(nlli_train)
#test_nlli = dcgan.evaluate_neg_loglikelihood(test_data[0, :], FLAGS)
#test_nlli = dcgan.evaluate_neg_loglikelihood(test_data[0, :].repeat(FLAGS.batch_size, 1), FLAGS)
#test_nlli = dcgan.evaluate_neg_loglikelihood(test_data[0, :].repeat(FLAGS.batch_size, 1), FLAGS)
test_nlli = dcgan.evaluate_neg_loglikelihood(
np.tile(test_data[0, :], (FLAGS.batch_size, 1)), FLAGS)
val_nlli = dcgan.evaluate_neg_loglikelihood(
np.tile(val_data[0, :], (FLAGS.batch_size, 1)), FLAGS)
train_nlli = dcgan.evaluate_neg_loglikelihood(
np.tile(train_data[0, :], (FLAGS.batch_size, 1)), FLAGS)
#val_nlli = dcgan.evaluate_neg_loglikelihood(val_data[0, :].repeat(FLAGS.batch_size, 1), FLAGS)
#train_nlli = dcgan.evaluate_neg_loglikelihood(train_data[0, :].repeat(FLAGS.batch_size, 1), FLAGS)
#print(test_nlli)
#print(test_nlli)
print('')
print(test_nlli)
print(val_nlli)
print(train_nlli)
#print(val_nlli)
#print(val_nlli)
#print(train_nlli)
#print(train_nlli)
#print(train_nlli)
#print(train_nlli)
#print(train_nlli)
#adsfadsf
#dsfasf
#adsfas
#print('')
#print('')
print('')
#print('')
#print('asdfaasdfsssasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdfaasdfsssasdfasdfassdasdfasfdfasf')
#print('asdfaassasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdadsfsasfsasdfsfssafasddfsaasdf')
#print('asdfaasdfsssasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdfaasdfsssasdfasdfassdasdfasfdfasf')
#print('')
#print('asdfaasdfsssasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdfaasdfsssasdfasdfassdasdfasfdfasf')
#print('asdfaasdfsssasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdfaasdfsssasdfasdfassdasdfasfdfasf')
#print('asdfaasdfsssasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdfaasdfsssasdfasdfassdasdfasfdfasf')
#print('asdfaassasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdadsfsasfsasdfsfssafasddfsaasdf')
#print('asdfaassasdfasdfassdasdfasfdfasf OK. This is OK. OK. asdadsfsasfsasdfsfssafasddfsaasdf')
# evaluate_neg_loglikelihood(self, data, config)
# use: evaluate_neg_loglikelihood(self, data, config)
#train_nlli = dcgan.evaluate_neg_loglikelihood(np.tile(train_data[0, :], (FLAGS.batch_size, 1)), FLAGS)
#train_nlli = dcgan.evaluate_neg_loglikelihood(np.tile(train_data[0, :], (FLAGS.batch_size, 1)), FLAGS)
#dcdcdcgan.train(FLAGS2)
#dcdcdcgan.train(FLAGS2, dcgan, FLAGS)
#dcdcdcgan.train(FLAGS2, dcgan, FLAGS)
dcdcdcgan.train(FLAGS2)
