AgentBase.py中std的计算是否有问题

[ljn114514]

第235行部分self.cri.state_std[:] = self.cri.state_std，这样的话cri.state_std是永远不变的

[Yonv1943]

谢谢你的提醒，我们检查了相关函数，的确发现问题，如下：

在函数 def update_avg_std_for_normalization 里，我们更新了 state_std：

ElegantRL/elegantrl/agents/AgentBase.py

Lines 230 to 240 in bf5ecff

	state_avg = states.mean(dim=0, keepdim=True)
	state_std = states.std(dim=0, keepdim=True)
	self.act.state_avg[:] = self.act.state_avg * (1 - tau) + state_avg * tau
	self.act.state_std[:] = self.cri.state_std * (1 - tau) + state_std * tau + 1e-4
	self.cri.state_avg[:] = self.act.state_avg
	self.cri.state_std[:] = self.cri.state_std
	returns_avg = returns.mean(dim=0)
	returns_std = returns.std(dim=0)
	self.cri.value_avg[:] = self.cri.value_avg * (1 - tau) + returns_avg * tau
	self.cri.value_std[:] = self.cri.value_std * (1 - tau) + returns_std * tau + 1e-4

应该改成

        self.cri.state_avg[:] = self.act.state_avg
        self.cri.state_std[:] = self.act.state_std   # 这里应该做出修改

[Yonv1943]

#304

我提交了一个PR修复您提及的bug，再次感谢你