def _val_or_test(self, dst):
""" common component for val and run
dst: save record to dst filename
config: config file to running
"""
# get data pipeline
data, label, path = loads(self.config)
x1, x2 = tf.unstack(data, axis=1)
path1, path2 = tf.unstack(path, axis=1)
# get loss
x1, x2 = self._network(x1, x2)
loss, loss_batch = cosine.get_loss(x1, x2, label, self.batchsize,
False)
# write records
info = string.concat(self.batchsize,
[path1, path1, path2, label, loss_batch])
# setting session running info
output = [x1, x2, label, info]
saver = tf.train.Saver()
with context.DefaultSession() as sess:
step = self.snapshot.restore(sess, saver)
fw = open('%s/%s.txt' % (dst, step), 'wb')
with context.QueueContext(sess):
for i in range(self.epoch_iter):
_f1, _f2, _label, _info = sess.run(output)
self._write_feat_to_npy(i, _f1, _f2, _label)
[fw.write(_line + b'\r\n') for _line in _info]
fw.close()
return step
def train(self):
# set phase
self._enter_('train')
# get data pipeline
data, info, _ = loads(self.config)
c1_real, p1_real, c2_real, p2_real = tf.unstack(data, axis=1)
label, cond = tf.unstack(info, axis=1)
# encode image to a vector
c1_mu, c1_sigma, feat_c1 = self._encoder(c1_real)
p2_mu, p2_sigma, feat_p2 = self._encoder(p2_real, True)
# children to parent
c1_z = c1_mu + c1_sigma * tf.random_normal(tf.shape(c1_mu))
c1_z = self._generator(c1_z, cond)
c1_fake = tf.clip_by_value(c1_z, 1e-8, 1 - 1e-8)
# parent to children
p2_z = p2_mu + p2_sigma * tf.random_normal(tf.shape(p2_mu))
p2_z = self._generator(p2_z, cond, True)
p2_fake = tf.clip_by_value(p2_z, 1e-8, 1 - 1e-8)
# loss for genertor
E1_loss = self._loss_vae(c1_real, c1_fake, c1_mu, c1_sigma)
E2_loss = self._loss_vae(p2_real, p2_fake, p2_mu, p2_sigma)
E_loss = E1_loss + E2_loss
loss = E_loss
# # allocate two optimizer
global_step = tf.train.create_global_step()
var_e = variables.select_vars('encoder')
var_g = variables.select_vars('generator')
op1 = updater.default(self.config, loss, global_step, var_e, 0)
op2 = updater.default(self.config, loss, None, var_g, 1)
train_op = tf.group(op1, op2)
# update at the same time
saver = tf.train.Saver(var_list=variables.all())
# hooks
self.add_hook(self.snapshot.init())
self.add_hook(self.summary.init())
self.add_hook(
context.Running_Hook(config=self.config.log,
step=global_step,
keys=['E'],
values=[E_loss],
func_test=self.test,
func_val=None))
with context.DefaultSession(self.hooks) as sess:
self.snapshot.restore(sess, saver)
while not sess.should_stop():
sess.run(train_op)
def train(self):
"""
"""
# set phase
self._enter_('train')
# get data pipeline
data, label, _ = loads(self.config)
# generate fake images
logit_G, net_G = self._generator(label)
# discriminate fake images
logit_F, net_F, out_F = self._discriminator(logit_G,
label,
reuse=False)
# discriminate real images
logit_R, net_R, out_R = self._discriminator(data, label, reuse=True)
# classify fake
c_F = self._classifier(out_F, reuse=False)
# classify real
c_R = self._classifier(out_R, reuse=True)
# loss
D_loss, G_loss, C_loss = self._loss(label, logit_F, logit_R, c_F, c_R)
# t_vars = tf.trainable_variables()
d_vars = variables.select_vars('discriminator')
g_vars = variables.select_vars('generator')
c_vars = variables.select_vars('classifier')
# update
step = tf.train.create_global_step()
d_op = updater.default(self.config, D_loss, step, d_vars, 0)
g_op = updater.default(self.config, G_loss, step, g_vars, 0)
c_op = updater.default(self.config, C_loss, step, c_vars, 0)
# assemble
train_op = [[d_op, g_op, c_op]]
saver = tf.train.Saver(variables.all())
# hooks
self.add_hook(self.snapshot.init())
self.add_hook(self.summary.init())
self.add_hook(
context.Running_Hook(config=self.config.log,
step=step,
keys=['D_loss', 'G_loss', 'C_loss'],
values=[D_loss, G_loss, C_loss],
func_test=self.test,
func_val=None))
# monitor session
with context.DefaultSession(self.hooks) as sess:
self.snapshot.restore(sess, saver)
while not sess.should_stop():
sess.run(train_op)
def test(self):
"""
"""
# save current context
self._enter_('test')
# create a folder to save
test_dir = filesystem.mkdir(self.config.output_dir + '/test/')
# get data pipeline
data, label, path = loads(self.config)
# get network
logit, net = self._net(data)
loss, mae, rmse = self._loss(logit, label)
# prepare
total_num = self.data.total_num
batchsize = self.data.batchsize
num_iter = int(total_num / batchsize)
info = string.concat(batchsize, [path, label, logit * self.data.range])
mean_loss, mean_mae, mean_rmse = 0, 0, 0
# get saver
saver = tf.train.Saver()
with tf.Session() as sess:
# get latest checkpoint
global_step = self.snapshot.restore(sess, saver)
with open(test_dir + '%s.txt' % global_step, 'wb') as fw:
with context.QueueContext(sess):
for _ in range(num_iter):
# running session to acuqire value
_loss, _mae, _rmse, _info = sess.run(
[loss, mae, rmse, info])
mean_loss += _loss
mean_mae += _mae
mean_rmse += _rmse
# save tensor info to text file
[fw.write(_line + b'\r\n') for _line in _info]
# statistic
mean_loss = 1.0 * mean_loss / num_iter
mean_mae = 1.0 * mean_mae / num_iter
mean_rmse = 1.0 * mean_rmse / num_iter
# display results on screen
keys = ['total sample', 'num batch', 'loss', 'mae', 'rmse']
vals = [total_num, num_iter, mean_loss, mean_mae, mean_rmse]
logger.test(logger.iters(int(global_step), keys, vals))
# write to summary
self.summary.adds(global_step=global_step,
tags=['test/loss', 'test/mae', 'test/rmse'],
values=[mean_loss, mean_mae, mean_rmse])
self._exit_()
return mean_mae
def test(self):
"""
"""
# save current context
self._enter_('test')
# create a folder to save
test_dir = filesystem.mkdir(self.config.output_dir + '/test/')
# get data pipeline
data, label, path = loads(self.config)
# total_num
total_num = self.data.total_num
batchsize = self.data.batchsize
# network
logit, net = self._net(data)
loss, error, pred = self._loss(logit, label)
# prepare
info = string.concat(batchsize, [path, label, pred])
num_iter = int(total_num / batchsize)
mean_err, mean_loss = 0, 0
# get saver
saver = tf.train.Saver()
with context.DefaultSession() as sess:
global_step = self.snapshot.restore(sess, saver)
with open(test_dir + '%s.txt' % global_step, 'wb') as fw:
with context.QueueContext(sess):
for _ in range(num_iter):
_loss, _err, _info = sess.run([loss, error, info])
mean_loss += _loss
mean_err += _err
[fw.write(_line + b'\r\n') for _line in _info]
# statistic
mean_loss = 1.0 * mean_loss / num_iter
mean_err = 1.0 * mean_err / num_iter
# display results on screen
keys = ['total sample', 'num batch', 'loss', 'error']
vals = [total_num, num_iter, mean_loss, mean_err]
logger.test(logger.iters(int(global_step), keys, vals))
# write to summary
self.summary.adds(global_step=global_step,
tags=['test/error', 'test/loss'],
values=[mean_err, mean_loss])
self._exit_()
return mean_err
def train(self):
"""
"""
# set phase
self._enter_('train')
# get data pipeline
data, labels, path = loads(self.config)
# data = tf.reduce_sum(data, axis=3)
# data = tf.expand_dims(data, axis=3)
# data = tf.sign(data)
label, cond = tf.unstack(labels, axis=1)
# get network
logit, net = self._net(data)
cond = tf.reshape(cond, [self.data.batchsize, 1])
label = tf.reshape(label, [self.data.batchsize, 1])
label = label * cond
# label = label * (cond - 234.) / (689. - 234.)
# get loss
loss, mae, rmse = self._loss(logit, label)
# update
global_step = tf.train.create_global_step()
train_op = updater.default(self.config, loss, global_step)
# for storage
saver = tf.train.Saver(var_list=variables.all())
# hooks
snapshot_hook = self.snapshot.init()
summary_hook = self.summary.init()
running_hook = context.Running_Hook(config=self.config.log,
step=global_step,
keys=['loss', 'mae', 'rmse'],
values=[loss, mae, rmse],
func_test=self.test,
func_val=None)
# monitor session
with tf.train.MonitoredTrainingSession(
hooks=[
running_hook, snapshot_hook, summary_hook,
tf.train.NanTensorHook(loss)
],
save_checkpoint_secs=None,
save_summaries_steps=None) as sess:
self.snapshot.restore(sess, saver)
while not sess.should_stop():
sess.run(train_op)
def _val_or_test(self, dstdir):
""" COMMON FOR TRAIN AND VAL """
# considering output train image
data, info, path = loads(self.config)
c1_real, p1_real, c2_real, p2_real = tf.unstack(data, axis=1)
c1_path, p1_path, c2_path, p2_path = tf.unstack(path, axis=1)
label, cond = tf.unstack(info, axis=1)
# encode image to a vector
feat_c1, feat_p2 = self._network(c1_real, p2_real)
R_loss, loss = self._loss_metric(feat_c1, feat_p2, None)
saver = tf.train.Saver()
c1_zs, p2_zs = 0, 0
with context.DefaultSession() as sess:
step = self.snapshot.restore(sess, saver)
info = string.concat(
self.batchsize,
[c1_path, p1_path, c2_path, p2_path, label, loss])
output = [info, feat_c1, feat_p2, label]
fw = open(dstdir + '%s.txt' % step, 'wb')
with context.QueueContext(sess):
for i in range(self.epoch_iter):
_info, _x, _y, _label = sess.run(output)
self._write_feat_to_npy(i, _x, _y, _label)
[fw.write(_line + b'\r\n') for _line in _info]
c1_zs = _x if type(c1_zs) == int else np.row_stack(
(c1_zs, _x))
p2_zs = _y if type(p2_zs) == int else np.row_stack(
(p2_zs, _y))
fw.close()
np.save(dstdir + '/' + step + '_c1.npy', c1_zs)
np.save(dstdir + '/' + step + '_p2.npy', p2_zs)
return step
def train(self):
"""
"""
""" """
# set phase
self._enter_('train')
# get data pipeline
data, label, path = loads(self.config)
x1, x2 = tf.unstack(data, axis=1)
path1, path2 = tf.unstack(path, axis=1)
# encode image to a vector
x1, x2 = self._network(x1, x2)
loss, loss_batch = cosine.get_loss(x1, x2, label, self.batchsize, True)
# # allocate two optimizer
global_step = tf.train.create_global_step()
train_op = updater.default(self.config, loss, global_step)
# update at the same time
saver = tf.train.Saver(var_list=variables.all())
# hooks
self.add_hook(self.snapshot.init())
self.add_hook(self.summary.init())
self.add_hook(
context.Running_Hook(config=self.config.log,
step=global_step,
keys=['R'],
values=[loss],
func_test=self.test,
func_val=None))
with context.DefaultSession(self.hooks) as sess:
self.snapshot.restore(sess, saver)
while not sess.should_stop():
sess.run(train_op)
def train(self):
"""
"""
# set phase
self._enter_('train')
# get data pipeline
data, label, path = loads(self.config)
# get network
logit, net = self._net(data)
# get loss
loss, error, pred = self._loss(logit, label)
# update
global_step = tf.train.create_global_step()
train_op = updater.default(self.config, loss, global_step)
# for storage
saver = tf.train.Saver(var_list=variables.all())
# hooks
self.add_hook(self.snapshot.init())
self.add_hook(self.summary.init())
self.add_hook(
context.Running_Hook(config=self.config.log,
step=global_step,
keys=['loss', 'error'],
values=[loss, error],
func_test=self.test,
func_val=None))
# monitor session
with context.DefaultSession(self.hooks) as sess:
self.snapshot.restore(sess, saver)
# Profiler.time_memory(self.config['output_dir'], sess, train_op)
while not sess.should_stop():
sess.run(train_op)
def train(self):
""" """
# set phase
self._enter_('train')
# get data pipeline
data, info, _ = loads(self.config)
c1_real, p1_real, c2_real, p2_real = tf.unstack(data, axis=1)
label, cond = tf.unstack(info, axis=1)
# encode image to a vector
feat_c1, feat_p2 = self._network(c1_real, p2_real)
loss, loss_batch = self._loss_metric(feat_c1, feat_p2, label)
# # allocate two optimizer
global_step = tf.train.create_global_step()
train_op = updater.default(self.config, loss, global_step)
# update at the same time
saver = tf.train.Saver(var_list=variables.all())
# hooks
self.add_hook(self.snapshot.init())
self.add_hook(self.summary.init())
self.add_hook(context.Running_Hook(
config=self.config.log,
step=global_step,
keys=['R'],
values=[loss],
func_test=self.test,
func_val=None))
with context.DefaultSession(self.hooks) as sess:
self.snapshot.restore(sess, saver)
while not sess.should_stop():
sess.run(train_op)
def train(self):
"""
"""
# set phase
self._enter_('train')
# get data pipeline
data, label, path = loads(self.config)
print(data, label, path)
# get network
logit, net = self._net(data)
# get loss
loss, mae, rmse = self._loss(logit, label)
# update
global_step = tf.train.create_global_step()
train_op = updater.default(self.config, loss, global_step)
# update at the same time
saver = tf.train.Saver(var_list=variables.all())
# hooks
self.add_hook(self.snapshot.init())
self.add_hook(self.summary.init())
self.add_hook(
context.Running_Hook(config=self.config.log,
step=global_step,
keys=['loss', 'mae', 'rmse'],
values=[loss, mae, rmse],
func_test=self.test,
func_val=None))
with context.DefaultSession(self.hooks) as sess:
self.snapshot.restore(sess, saver)
while not sess.should_stop():
sess.run(train_op)
def _val_or_test(self, dstdir):
""" COMMON FOR TRAIN AND VAL """
# considering output train image
data, info, path = loads(self.config)
c1_real, p1_real, c2_real, p2_real = tf.unstack(data, axis=1)
c1_path, p1_path, c2_path, p2_path = tf.unstack(path, axis=1)
label, cond = tf.unstack(info, axis=1)
# encode image to a vector
c1_mu, c1_sigma, feat_c1 = self._encoder(c1_real)
p2_mu, p2_sigma, feat_p2 = self._encoder(p2_real, True)
c1_z = c1_mu + c1_sigma * tf.random_normal(tf.shape(c1_mu))
c1_z = self._generator(c1_z, cond)
c1_fake = tf.clip_by_value(c1_z, 1e-8, 1 - 1e-8)
p2_z = p2_mu + p2_sigma * tf.random_normal(tf.shape(p2_mu))
p2_z = self._generator(p2_z, cond, True)
p2_fake = tf.clip_by_value(p2_z, 1e-8, 1 - 1e-8)
R_loss, loss = self._loss_metric(feat_c1, feat_p2, None)
saver = tf.train.Saver()
c1_zs, p2_zs = 0, 0
with context.DefaultSession() as sess:
step = self.snapshot.restore(sess, saver)
filesystem.mkdir(dstdir + '/' + step + '_c1')
filesystem.mkdir(dstdir + '/' + step + '_p2')
info = string.concat(
self.data.batchsize,
[c1_path, p1_path, c2_path, p2_path, label, loss])
output = [
c1_fake, p2_fake, c1_path, p2_path, c1_real, c2_real, p1_real,
p2_real, info, feat_c1, feat_p2, label
]
fw = open(dstdir + '%s.txt' % step, 'wb')
with context.QueueContext(sess):
for i in range(self.epoch_iter):
_c1, _p2, _c1p, _p2p, _c1r, _c2r, _p1r, _p2r, _info, _x, _y, _label = sess.run(
output)
self._write_feat_to_npy(i, _x, _y, _label)
[fw.write(_line + b'\r\n') for _line in _info]
if self.is_save_all_images:
image.save_multi(dstdir + '/' + step + '_c1', _c1,
_c1p)
image.save_multi(dstdir + '/' + step + '_p2', _p2,
_p2p)
if self.is_save_batch_images:
image.save_batchs_to_one(
imgs_list=[_c1, _p2, _c1r, _c2r, _p1r, _p2r],
batchsize=self.data.batchsize,
dst=dstdir,
step=step,
name_list=[
'_c1', '_p2', '_c1r', '_c2r', '_p1r', '_p2r'
])
if self.is_save_feats:
c1_zs = _x if type(c1_zs) == int else np.row_stack(
(c1_zs, _x))
p2_zs = _y if type(p2_zs) == int else np.row_stack(
(p2_zs, _y))
fw.close()
if self.is_save_feats:
np.save(dstdir + '/' + step + '_c1.npy', c1_zs)
np.save(dstdir + '/' + step + '_p2.npy', p2_zs)
return step
def train(self):
""" """
# set phase
self._enter_('train')
# get data pipeline
data, info, _ = loads(self.config)
c1_real, p1_real, c2_real, p2_real = tf.unstack(data, axis=1)
label, cond = tf.unstack(info, axis=1)
# encode image to a vector
c1_mu, c1_sigma, feat_c1 = self._encoder(c1_real)
p2_mu, p2_sigma, feat_p2 = self._encoder(p2_real, True)
# children to parent
with tf.variable_scope('net1'):
c1_z = c1_mu + c1_sigma * tf.random_normal(tf.shape(c1_mu))
c1_z = self._generator(c1_z, cond)
c1_fake = tf.clip_by_value(c1_z, 1e-8, 1 - 1e-8)
# parent to children
with tf.variable_scope('net2'):
p2_z = p2_mu + p2_sigma * tf.random_normal(tf.shape(p2_mu))
p2_z = self._generator(p2_z, cond)
p2_fake = tf.clip_by_value(p2_z, 1e-8, 1 - 1e-8)
# discriminator
D_c1_fake = self._discriminator(c1_fake, cond, reuse=False)
D_p1_real = self._discriminator(p1_real, cond, reuse=True)
D_p2_fake = self._discriminator(p2_fake, cond, reuse=True)
D_c2_real = self._discriminator(c2_real, cond, reuse=True)
# loss for encoder
R_loss, _ = self._loss_metric(feat_c1, feat_p2, label)
# loss for genertor
E1_loss = self._loss_vae(p1_real, c1_fake, c1_mu, c1_sigma)
E2_loss = self._loss_vae(c2_real, p2_fake, p2_mu, p2_sigma)
E_loss = E1_loss + E2_loss
# loss for discriminator
D1_loss, G1_loss = self._loss_gan(D_c1_fake, D_p1_real)
D2_loss, G2_loss = self._loss_gan(D_p2_fake, D_c2_real)
D_loss = D1_loss + D2_loss
G_loss = G1_loss + G2_loss
loss = E_loss + D_loss + G_loss + R_loss
# # allocate two optimizer
global_step = tf.train.create_global_step()
var_e = variables.select_vars('encoder')
var_g = variables.select_vars('generator')
var_d = variables.select_vars('discriminator')
op1 = updater.default(self.config, loss, global_step, var_e, 0)
op2 = updater.default(self.config, loss, None, var_g, 1)
op3 = updater.default(self.config, loss, None, var_d, 0)
train_op = tf.group(op1, op2, op3)
# update at the same time
saver = tf.train.Saver(var_list=variables.all())
# hooks
self.add_hook(self.snapshot.init())
self.add_hook(self.summary.init())
self.add_hook(
context.Running_Hook(config=self.config.log,
step=global_step,
keys=['E', 'D', 'G', 'R'],
values=[E_loss, D_loss, G_loss, R_loss],
func_test=self.test,
func_val=None))
with context.DefaultSession(self.hooks) as sess:
self.snapshot.restore(sess, saver)
while not sess.should_stop():
sess.run(train_op)
