moving epsilon outside softplus

I think you intended this to be outside the softplus. The reason is that it should be applied just before the log to avoid `log(0)=inf`.e.g.

- `log(softplus(-1000+1e-5))=log(0)=inf`. 
- `log(softplus(-1000)+1e-5)=log(1e-5)!=inf`. 

Also this fixes a NaN I had.

network/continuous_action_network.py

@@ -145,7 +145,7 @@ class GaussianActorNet(nn.Module, BasicNet):
             log_std = self.action_log_std.expand_as(mean)
             std = log_std.exp()
         else:
-            std = F.softplus(self.action_std(phi) + 1e-5)
+            std = F.softplus(self.action_std(phi)) + 1e-5
             log_std = std.log()
         return mean, std, log_std