def get_variables_with_name(name, train_only=True, printable=False):
"""Get variable list by a given name scope.
Examples
---------
>>> dense_vars = tl.layers.get_variable_with_name('dense', True, True)
"""
print(" [*] geting variables with %s" % name)
# tvar = tf.trainable_variables() if train_only else tf.all_variables()
if train_only:
t_vars = tf.trainable_variables()
else:
try: # TF1.0
t_vars = tf.global_variables()
except: # TF0.12
t_vars = tf.all_variables()
d_vars = [var for var in t_vars if name in var.name]
if printable:
for idx, v in enumerate(d_vars):
print(" got {:3}: {:15} {}".format(idx, v.name, str(v.get_shape())))
return d_vars
python类all_variables()的实例源码
def decode():
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model_default(sess, forward_only=True, attention=attention, model_path=model_path)
#for v in tf.all_variables():
# print(v.name, v.get_shape())
# eval_batch_size = 64
# start_time = time.time()
# do_evalb(model_dev, sess, dev_set, eval_batch_size)
# time_elapsed = time.time() - start_time
# print("Batched evalb time: ", time_elapsed)
start_time = time.time()
write_decode(model_dev, sess, dev_set)
time_elapsed = time.time() - start_time
print("Decoding all dev time: ", time_elapsed)
def decode(debug=True):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model(sess, forward_only=True, dropout=False, model_path=FLAGS.model_path)
if debug:
for v in tf.all_variables(): print(v.name, v.get_shape())
eval_batch_size = 64
start_time = time.time()
do_evalb(model_dev, sess, dev_set, eval_batch_size)
time_elapsed = time.time() - start_time
print("Batched evalb time: ", time_elapsed)
# start_time = time.time()
# write_decode(model_dev, sess, dev_set, eval_batch_size)
# time_elapsed = time.time() - start_time
# print("Decoding all dev time: ", time_elapsed)
def decode():
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model_default(sess, forward_only=True, dropout=False, model_path=model_path)
#for v in tf.all_variables():
# print(v.name, v.get_shape())
# eval_batch_size = 64
# start_time = time.time()
# do_evalb(model_dev, sess, dev_set, eval_batch_size)
# time_elapsed = time.time() - start_time
# print("Batched evalb time: ", time_elapsed)
start_time = time.time()
write_decode(model_dev, sess, dev_set)
time_elapsed = time.time() - start_time
print("Decoding all dev time: ", time_elapsed)
def save_vars(filename):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model_default(sess, forward_only=True, dropout=False, model_path=model_path)
var_dict = {}
for var in tf.all_variables():
if 'Adagrad' in var.name: continue
var_dict[var.name] = var.eval()
pickle.dump(var_dict, open(filename, 'w'))
#for v in tf.all_variables():
# print(v.name, v.get_shape())
def decode(debug=True):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model(sess, forward_only=True, dropout=False, model_path=FLAGS.model_path)
if debug:
for v in tf.all_variables(): print(v.name, v.get_shape())
eval_batch_size = 64
start_time = time.time()
do_evalb(model_dev, sess, dev_set, eval_batch_size)
time_elapsed = time.time() - start_time
print("Batched evalb time: ", time_elapsed)
# start_time = time.time()
# write_decode(model_dev, sess, dev_set, eval_batch_size)
# time_elapsed = time.time() - start_time
# print("Decoding all dev time: ", time_elapsed)
def decode(debug=True):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model(sess, forward_only=True, dropout=False, model_path=model_path)
if debug:
for v in tf.all_variables(): print(v.name, v.get_shape())
eval_batch_size = 64
start_time = time.time()
do_evalb(model_dev, sess, dev_set, eval_batch_size)
time_elapsed = time.time() - start_time
print("Batched evalb time: ", time_elapsed)
# start_time = time.time()
# write_decode(model_dev, sess, dev_set, eval_batch_size)
# time_elapsed = time.time() - start_time
# print("Decoding all dev time: ", time_elapsed)
def decode(debug=True):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model(sess, forward_only=True, dropout=False, model_path=model_path)
if debug:
for v in tf.all_variables(): print(v.name, v.get_shape())
eval_batch_size = 64
start_time = time.time()
do_evalb(model_dev, sess, dev_set, eval_batch_size)
time_elapsed = time.time() - start_time
print("Batched evalb time: ", time_elapsed)
# start_time = time.time()
# write_decode(model_dev, sess, dev_set, eval_batch_size)
# time_elapsed = time.time() - start_time
# print("Decoding all dev time: ", time_elapsed)
def decode(debug=True):
""" Decode file """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model(sess, forward_only=True)
if debug:
for v in tf.all_variables(): print(v.name, v.get_shape())
dev_set = load_dev_data()
eval_batch_size = FLAGS.batch_size
start_time = time.time()
write_decode(model_dev, sess, dev_set, eval_batch_size, steps_done, eval_now=True)
time_elapsed = time.time() - start_time
print("Decoding all dev time: ", time_elapsed)
def decode(debug=True):
""" Decode file """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model(sess, forward_only=True)
if debug:
for v in tf.all_variables(): print(v.name, v.get_shape())
dev_set = load_dev_data()
eval_batch_size = FLAGS.batch_size
start_time = time.time()
write_decode(model_dev, sess, dev_set, eval_batch_size, steps_done, eval_now=True)
time_elapsed = time.time() - start_time
print("Decoding all dev time: ", time_elapsed)
def decode(debug=True):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model(sess, forward_only=True, dropout=False, model_path=FLAGS.model_path)
if debug:
for v in tf.all_variables(): print(v.name, v.get_shape())
eval_batch_size = 64
start_time = time.time()
do_evalb(model_dev, sess, dev_set, eval_batch_size)
time_elapsed = time.time() - start_time
print("Batched evalb time: ", time_elapsed)
# start_time = time.time()
# write_decode(model_dev, sess, dev_set, eval_batch_size)
# time_elapsed = time.time() - start_time
# print("Decoding all dev time: ", time_elapsed)
def save_vars(filename):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model_default(sess, forward_only=True, dropout=False, model_path=model_path)
var_dict = {}
for var in tf.all_variables():
print(var.name, var.get_shape())
if 'Adagrad' in var.name: continue
var_dict[var.name] = var.eval()
pickle.dump(var_dict, open(filename, 'w'))
#for v in tf.all_variables():
# print(v.name, v.get_shape())
def decode():
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model_default(sess, forward_only=True, attention=attention, model_path=model_path)
#with tf.variable_scope("model", reuse=True):
# model_dev = seq2seq_model.Seq2SeqModel(90000, 128, _buckets, 256, 3, 512, 5.0, 128, 0.1, 0.99, forward_only=True, attention=attention)
#for v in tf.all_variables():
# print(v.name, v.get_shape())
#eval_batch_size = 64
#start_time = time.time()
#do_evalb(model_dev, sess, dev_set, eval_batch_size)
#time_elapsed = time.time() - start_time
#print("Batched evalb time: ", time_elapsed)
start_time = time.time()
write_decode(model_dev, sess, dev_set)
time_elapsed = time.time() - start_time
print("Decoding all dev time: ", time_elapsed)
def decode():
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model_default(sess, forward_only=True, attention=attention, model_path=model_path)
# for v in tf.all_variables():
# print(v.name, v.get_shape())
eval_batch_size = 64
start_time = time.time()
do_evalb(model_dev, sess, dev_set, eval_batch_size)
time_elapsed = time.time() - start_time
print("Batched evalb time: ", time_elapsed)
# start_time = time.time()
# write_decode(model_dev, sess, dev_set)
# time_elapsed = time.time() - start_time
# print("Decoding all dev time: ", time_elapsed)
def save_vars(filename):
""" Decode file sentence-by-sentence """
with tf.Session(config=tf.ConfigProto(intra_op_parallelism_threads=NUM_THREADS)) as sess:
# Create model and load parameters.
with tf.variable_scope("model", reuse=None):
model_dev, steps_done = create_model_default(sess, forward_only=True, dropout=False, model_path=model_path)
var_dict = {}
for var in tf.all_variables():
if 'Adagrad' in var.name: continue
var_dict[var.name] = var.eval()
pickle.dump(var_dict, open(filename, 'w'))
#for v in tf.all_variables():
# print(v.name, v.get_shape())
def main(_):
"""
Main function, loads and vectorizes the data, instantiates the network variables, and runs the
training process.
"""
# Perform data preprocessing here
# TODO
with tf.Session() as sess:
# Instantiate Network
print 'Building Network!'
rlangmod = RLangmod(FLAGS.vocabulary_size, FLAGS.embedding_size, FLAGS.hidden_size)
# Create a saver.
saver = tf.train.Saver(tf.all_variables())
# Initialize all variables
print "Initializing Variables"
sess.run(tf.initialize_all_variables())
def save_h5(args, net):
# Begin tf session
with tf.Session() as sess:
tf.initialize_all_variables().run()
saver = tf.train.Saver(tf.all_variables())
# load from previous save
if len(args.ckpt_name) > 0:
saver.restore(sess, os.path.join(args.save_dir, args.ckpt_name))
else:
print 'checkpoint name not specified... exiting.'
return
vs = tf.get_collection(tf.GraphKeys.VARIABLES)
vals = sess.run(vs)
exclude = ['learning_rate', 'beta', 'Adam']
with h5py.File(args.h5_name, 'a') as f:
dset = f.create_group('iter00001')
for v, val in safezip(vs, vals):
if all([e not in v.name for e in exclude]):
dset[v.name] = val
# Train network
def set_optimizer(self, session, learning_rate=0.5, learning_rate_decay_factor=0.99, max_gradient_norm=5.0, load_if_exist=True):
self.global_step = tf.Variable(0, trainable=False)
self.learning_rate = tf.Variable(float(learning_rate), trainable=False)
self.learning_rate_opr = self.learning_rate.assign(self.learning_rate * learning_rate_decay_factor)
self.optimizer = tf.train.GradientDescentOptimizer(self.learning_rate)
self.outputs, self.losses = self.calc_loss()
params = tf.trainable_variables()
for b in range(len(self.buckets)):
gradients = tf.gradients(self.losses[b], params)
clipped_gradients, norm = tf.clip_by_global_norm(gradients, max_gradient_norm)
self.gradient_norms.append(norm)
self.updates.append(self.optimizer.apply_gradients(zip(clipped_gradients, params), global_step=self.global_step))
self.saver = tf.train.Saver(tf.all_variables())
session.run(tf.initialize_all_variables())
if load_if_exist and self.train_dir:
saved = tf.train.get_checkpoint_state(self.train_dir)
if saved and tf.gfile.Exists(saved.model_checkpoint_path):
self.saver.restore(session, saved.model_checkpoint_path)
def build(self, session, predict=True, projection=True):
for j, bucket in enumerate(self.buckets):
with vs.variable_scope(vs.get_variable_scope(), reuse=True if j > 0 else None):
o, d_s, e_s = self.model.forward(
self.encoder_inputs[:bucket[0]], self.decoder_inputs[:bucket[1]], predict=predict, projection=projection
)
self._outputs.append(o)
self._encoder_state.append(e_s)
self._decoder_state.append(d_s)
self.saver = tf.train.Saver(tf.all_variables())
session.run(tf.initialize_all_variables())
if self.model_path:
saved = tf.train.get_checkpoint_state(self.model_path)
if saved and tf.gfile.Exists(saved.model_checkpoint_path):
self.saver.restore(session, saved.model_checkpoint_path)
self._graph_builded = True
def set_optimizer(self, session, learning_rate=0.1, learning_rate_decay_factor=0.99, max_gradient_norm=5.0, load_if_exist=True):
self.global_step = tf.Variable(0, trainable=False)
self.learning_rate = tf.Variable(float(learning_rate), trainable=False)
self.learning_rate_opr = self.learning_rate.assign(self.learning_rate * learning_rate_decay_factor)
self.optimizer = tf.train.GradientDescentOptimizer(self.learning_rate)
self.outputs, self.losses = self.calc_loss()
params = tf.trainable_variables()
for b in range(len(self.buckets)):
gradients = tf.gradients(self.losses[b], params)
clipped_gradients, norm = tf.clip_by_global_norm(gradients, max_gradient_norm)
self.gradient_norms.append(norm)
self.updates.append(self.optimizer.apply_gradients(zip(clipped_gradients, params), global_step=self.global_step))
self.saver = tf.train.Saver(tf.all_variables())
session.run(tf.initialize_all_variables())
if load_if_exist and self.train_dir:
saved = tf.train.get_checkpoint_state(self.train_dir)
if saved and tf.gfile.Exists(saved.model_checkpoint_path):
self.saver.restore(session, saved.model_checkpoint_path)
