61423919aac550098ab529f70a744af7c5ec0dbe,chainerrl/agents/a2c.py,A2C,batch_observe_and_train,#A2C#Any#Any#Any#Any#,260

Before Change

    def batch_observe_and_train(self, batch_obs, batch_reward, batch_done,
                                batch_reset):

        if any(batch_reset):
            raise RuntimeError("A2C does not support resetting an env withtout reaching a terminal state during training")  // NOQA

        statevar = self.batch_states(batch_obs, self.xp, self.phi)

        self.masks[self.t - self.t_start - 1] =\
            self.xp.array([0.0 if done else 1.0 for done in batch_done])

After Change

        if any(batch_reset):
            warnings.warn("A2C currently does not support resetting an env withtout reaching a terminal state during training. When receiving True in batch_reset, A2C considers it as True in batch_done instead.")  // NOQA
            for i, reset in enumerate(batch_reset):
                if reset:
                    batch_done[i] = True

        statevar = self.batch_states(batch_obs, self.xp, self.phi)

        self.masks[self.t - self.t_start - 1] =\
            self.xp.array([0.0 if done else 1.0 for done in batch_done])

Instances