def build_training_process(self):
c = tf.constant(0.8, dtype=tf.float32)
# if self.wider_seg_deepr > 0, then get wide_side_obj, else wider_entropy = 0
wider_side_obj, wider_entropy = tf.cond(
tf.greater(self.wider_seg_deeper,
0), lambda: self.get_wider_side_obj(), lambda:
(tf.constant(0.0, dtype=tf.float32),
tf.constant(0.0, dtype=tf.float32)))
wider_side_obj = tf.reduce_sum(wider_side_obj *
tf.minimum(c, self.wider_rho))
batch_size = array_ops.shape(self.reward)[0]
deeper_side_obj, deeper_entropy = tf.cond(
self.has_deeper, lambda: self.get_deeper_side_obj(), lambda:
(tf.constant(0.0, dtype=tf.float32),
tf.constant(0.0, dtype=tf.float32)))
deeper_side_obj = tf.reduce_sum(deeper_side_obj *
tf.minimum(c, self.deeper_rho))
self.obj = wider_side_obj + deeper_side_obj
entropy_term = wider_entropy * tf.cast(self.wider_seg_deeper, tf.float32) + \
deeper_entropy * tf.cast(batch_size - self.wider_seg_deeper, tf.float32)
entropy_term /= tf.cast(batch_size, tf.float32)
g = -self.obj - self.entropy_penalty * entropy_term
optimizer = BasicModel.build_optimizer(self.learning_rate,
self.opt_config[0],
self.opt_config[1])
self.train_step = [optimizer.minimize(g)]
wq = tf.reshape(self.wider_actor.selected_q, [-1])
w_loss = tf.losses.mean_squared_error(self.wider_qrets, wq)
self.update_wider_q = tf.train.AdamOptimizer(
learning_rate=self.learning_rate).minimize(loss=w_loss)
dq = tf.reshape(self.deeper_actor.selected_q, [-1])
d_loss = tf.losses.mean_squared_error(self.deeper_qrets, dq)
self.update_deeper_q = tf.train.AdamOptimizer(
learning_rate=self.learning_rate).minimize(loss=d_loss)
# add baseline to training step
if self.baseline_actor is not None:
self.build_baseline_network()
def build_training_process(self):
wider_side_obj, wider_entropy = tf.cond(
tf.greater(self.wider_seg_deeper,
0), lambda: self.get_wider_side_obj(), lambda:
(tf.constant(0.0, dtype=tf.float32),
tf.constant(0.0, dtype=tf.float32)))
batch_size = array_ops.shape(self.reward)[0]
deeper_side_obj, deeper_entropy = tf.cond(
self.has_deeper, lambda: self.get_deeper_side_obj(), lambda:
(tf.constant(0.0, dtype=tf.float32),
tf.constant(0.0, dtype=tf.float32)))
self.obj = wider_side_obj + deeper_side_obj
entropy_term = wider_entropy * tf.cast(self.wider_seg_deeper, tf.float32) + \
deeper_entropy * tf.cast(batch_size - self.wider_seg_deeper, tf.float32)
entropy_term /= tf.cast(batch_size, tf.float32)
optimizer = BasicModel.build_optimizer(self.learning_rate,
self.opt_config[0],
self.opt_config[1])
# print "in build, reward = {}".format(self.reward)
self.train_step = optimizer.minimize(-self.obj - self.entropy_penalty *
entropy_term)
def build_training_process(self):
# if self.wider_seg_deepr > 0, then get wide_side_obj, else wider_entropy = 0
wider_side_obj, wider_entropy = tf.cond(
tf.greater(self.wider_seg_deeper, 0),
lambda: self.get_wider_side_obj(),
lambda: (tf.constant(0.0, dtype=tf.float32), tf.constant(0.0, dtype=tf.float32))
)
batch_size = array_ops.shape(self.reward)[0]
deeper_side_obj, deeper_entropy = tf.cond(
self.has_deeper,
lambda: self.get_deeper_side_obj(),
lambda: (tf.constant(0.0, dtype=tf.float32), tf.constant(0.0, dtype=tf.float32))
)
self.obj = wider_side_obj + deeper_side_obj
entropy_term = wider_entropy * tf.cast(self.wider_seg_deeper, tf.float32) + \
deeper_entropy * tf.cast(batch_size - self.wider_seg_deeper, tf.float32)
entropy_term /= tf.cast(batch_size, tf.float32)
optimizer = BasicModel.build_optimizer(self.learning_rate, self.opt_config[0], self.opt_config[1])
self.train_step = [optimizer.minimize(- self.obj - self.entropy_penalty * entropy_term)]
# add baseline to training step
if self.baseline_actor is not None:
self.build_baseline_network()
