Noise added to predicted action during BC

[elicassion]

Hi,

In your code, it seems that you didn't remove noise added to your BatchNormMLP policy output. Is it intended or a bug? I'm confused since it's uncommon to have noise for BC policies.

https://github.com/aravindr93/mjrl/blob/83d35df95eb64274c5e93bb32a0a4e2f6576638a/mjrl/policies/gaussian_mlp.py#L90-L97

def get_action(self, observation):
        o = np.float32(observation.reshape(1, -1))
        self.obs_var.data = torch.from_numpy(o)
        mean = self.model(self.obs_var).data.numpy().ravel()
        noise = np.exp(self.log_std_val) * np.random.randn(self.m)
        action = mean + noise
        return [action, {'mean': mean, 'log_std': self.log_std_val, 'evaluation': mean}]