Weight features during data prep

Hello there,

I was wondering if a network is able to recognize small difference within a feature.
Let’s assume this feature is the only feature for a binary problem, it’s the numeric representation of occurrence of this feature in the original data.
To be labeled as A it needs to be 0 everything else would be B.
But also B just ranges from 1 to 5 for example. So the difference between A and B
could just be 1. Is a differentiation between these two still possible or would it be better
to put a weight to every occurrence of this feature so that there is a more distinct difference between the two ?