def update_policy(self, log_probs, cur_r, memory):
'''Imitate the poicy update of the learner'''
if self.alg == "vpg":
vpg_update(self.im_optimizer, log_probs, cur_r, memory.is_terminals, self.gamma)
elif self.alg == "ppo":
ppo_update(self.im_policy, self.im_optimizer, log_probs, cur_r, memory,
self.gamma, self.K_epochs, self.eps_clip, self.loss_fn, self.device)
def learning(self, memory):
if self.alg == "vpg":
old_states = torch.stack(memory.states).to(self.device).detach()
old_actions = torch.stack(memory.actions).to(self.device).detach()
logprobs = self.policy.act_prob(old_states, old_actions, self.device)
vpg_update(self.optimizer, logprobs, memory.rewards, memory.is_terminals, self.gamma)
elif self.alg == "ppo":
old_states = torch.stack(memory.states).to(self.device).detach()
old_actions = torch.stack(memory.actions).to(self.device).detach()
logprobs = self.policy.act_prob(old_states, old_actions, self.device)
ppo_update(self.policy, self.optimizer, logprobs, memory.rewards,
memory, self.gamma, self.K_epochs, self.eps_clip, self.loss_fn, self.device)