780dcd9fd372afa8524a6515eec6a4c90b1494c9,Reinforcement_learning_TUT/8_Actor_Critic_Advantage/AC_CartPole.py,Actor,__init__,#Actor#Any#Any#Any#,22

Before Change

class Actor(object):
    def __init__(self, n_features, n_actions, lr=0.001):
        with tf.name_scope("inputs"):
            self.state = tf.placeholder(tf.float32, [n_features, ], "state")
            state = tf.expand_dims(self.state, axis=0)
            self.act_index = tf.placeholder(tf.int32, name="act")
            self.advantage = tf.placeholder(tf.float32, name="adv")  // TD_error

        with tf.variable_scope("Actor"):
            l1 = tf.layers.dense(
                inputs=state,
                units=20,    // number of hidden units

After Change

        )

        with tf.variable_scope("squared_TD_error"):
            self.td_error = tf.reduce_mean(self.r + GAMMA * self.v_next - self.v)
            self.loss = tf.square(self.td_error)    // TD_error = (r+gamma*V_next) - V_eval
        with tf.variable_scope("train"):
            self.train_op = tf.train.AdamOptimizer(lr).minimize(self.loss)

Instances