16a31e2c9fedc654e9117b42b8144adf1d0e4900,examples/reinforcement_learning/tutorial_TRPO.py,TRPO,update,#TRPO#,633

Before Change

        // Prepare hessian func, gradient eval
        inputs = self.buf.get()
        Hx = lambda x: self.hvp(inputs, x)
        g, pi_l_old, v_l_old = self.gradient(inputs), self.pi_loss(inputs), self.v_loss(inputs)

        // Core calculations for TRPO or NPG
        x = self.cg(Hx, g)

After Change

        :return: None
        
        states, actions, adv, rewards_to_go, logp_old_ph, old_mu, old_log_std = self.buf.get()
        g, pi_l_old = self.gradient(states, actions, adv, logp_old_ph)

        Hx = lambda x: self.hvp(states, old_mu, old_log_std, x)
        x = self.cg(Hx, g)

Instances