ppo.py 文件源码

def _act(self, state):
        xp = self.xp
        with chainer.using_config('train', False):
            b_state = batch_states([state], xp, self.phi)
            with chainer.no_backprop_mode():
                action_distrib, v = self.model(b_state)
                action = action_distrib.sample()
            return cuda.to_cpu(action.data)[0], cuda.to_cpu(v.data)[0]