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

Before Change

            )

        with tf.name_scope("loss"):
            neg_log_prob = -tf.log(self.acts_prob[0, self.act_index])   // loss without advantage
            self.loss = tf.reduce_mean(neg_log_prob * self.advantage)  // advantage (TD_error) guided loss

        with tf.name_scope("train"):

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