373ae159f7ae1cabaf87228d1ae0fb6acd1c6363,ch14/lib/model.py,AgentDDPG,__call__,#AgentDDPG#,131

Before Change

        return None

    def __call__(self, states, agent_states):
        states_v = ptan.agent.float32_preprocessor(states, cuda=self.cuda)
        mu_v = self.net(states_v)
        actions = mu_v.data.cpu().numpy()

        if self.ou_enabled and self.ou_epsilon > 0:
            new_a_states = []
            for a_state, action in zip(agent_states, actions):
                if a_state is None:
                    a_state = np.zeros(shape=action.shape, dtype=np.float32)
                a_state += self.ou_teta * (self.ou_mu - a_state)
                a_state += self.ou_sigma * np.random.normal(size=action.shape)

                action += self.ou_epsilon * a_state

After Change

        return None

    def __call__(self, states, agent_states):
        states_v = ptan.agent.float32_preprocessor(states).to(self.device)
        mu_v = self.net(states_v)
        actions = mu_v.data.cpu().numpy()

        if self.ou_enabled and self.ou_epsilon > 0:
            new_a_states = []
            for a_state, action in zip(agent_states, actions):
                if a_state is None:
                    a_state = np.zeros(shape=action.shape, dtype=np.float32)
                a_state += self.ou_teta * (self.ou_mu - a_state)
                a_state += self.ou_sigma * np.random.normal(size=action.shape)

                action += self.ou_epsilon * a_state

Instances