def create ( cls, p_opcode, p_args ): #---------------------------------------------------------------------- l_format = opcodes.result_format(p_opcode) if l_format == None: return None l_data = data.pack(l_format, p_args) if l_data == None: return None return cls(p_opcode, l_data)
def create ( cls, p_opcode, p_args, p_timeout = 0.0, p_guid = None ): #---------------------------------------------------------------------- l_format = opcodes.command_format(p_opcode) if l_format == None: return None l_data = data.pack(l_format, p_args) if l_data == None: return None l_command = cls(p_opcode, l_data) l_command.set_timeout(p_timeout) l_command.set_guid(p_guid) return l_command
sess.run(tf.global_variables_initializer())
# tensor board
merged = tf.summary.merge_all()
writer = tf.summary.FileWriter(log_path + '/', sess.graph)
logging.info('Model builded, %s used\n' %
time_format(time.time() - tic0))
test_feed_dicts = []
for i in xrange((len(dataset_u.test_x) + args.batch_size_u - 1) //
args.batch_size_u):
x, y, l, msl = data.pack(
dataset_u.test_x[i * args.batch_size_u:(i + 1) *
args.batch_size_u],
dataset_u.test_y[i * args.batch_size_u:(i + 1) *
args.batch_size_u])
test_feed_dicts.append({
x_u_: x,
seq_len_u_: l,
y_u_: y,
msl_u_: msl,
keep_prob_: 1
})
train_s_feed_dicts = []
for i in xrange((len(dataset_s.train_x) + args.batch_size_s - 1) //
args.batch_size_s):
x, y, l, msl = data.pack(
dataset_s.train_x[i * args.batch_size_s:(i + 1) *
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
init = tf.initialize_all_variables()
sess.run(init)
logging.info('Model builded, %s used\n' % time_format(time.time() - tic0))
if not args.quiet:
print('Model builded, %s used\n' % time_format(time.time() - tic0))
# pre-store test dataset feed_dicts
test_feed_dicts = []
for i in xrange((len(dataset.test_x)+args.batch_size-1) // args.batch_size):
x, y, l, msl = data.pack(
dataset.test_x[i*args.batch_size:(i+1)*args.batch_size],
dataset.test_y[i*args.batch_size:(i+1)*args.batch_size],
args.window_size)
test_feed_dicts.append({x_:x, seq_len_:l, y_:y, msl_:msl, keep_prob_:1.})
for epoch in xrange(1, args.n_epochs+1):
tic = time.time()
loss = 0.
n_train = dataset.n_train
n_trained = 0
for idxs in dataset.minibatches:
x, y, l, msl = dataset.next_batch()
_, c = sess.run([train_op, loss_], feed_dict={x_:x, seq_len_:l, y_:y, msl_:msl, keep_prob_:args.keep_prob})
if np.isnan(c):
logging.error('Gradient Explosion!')
print('Gradient Explosion!')
exit()
