f2328f664ba2ad19868d6e0eaff08721d96c506b,tensorforce/models/pg_model.py,PGModel,tf_reward_estimation,#PGModel#Any#Any#Any#Any#Any#,161

Before Change

                )

            // Normalize advantage.
            mean, variance = tf.nn.moments(x=advantage, axes=[0], keep_dims=True)
            advantage = (advantage - mean) / tf.sqrt(x=tf.maximum(x=variance, y=util.epsilon))

            return advantage

After Change

                    update=update
                )
                state_value = self.baseline.predict(
                    states=tf.stop_gradient(input=embedding),
                    internals=internals,
                    update=update
                )

Instances