Feature Filtering

I’m training a recurrent net with 31 features which can be dived into 3 categories (20+6+5) , in some example it’s better to focus the 20 first features not on the 6 next and in some other it’s better to focus on the 6 over the first 20 , is there a way to tell to neural net of this depending on some extra condition/variable, I can add an extra 32th features in a binary form to address the issue.
And Thank you.