Support multiple position array normalizations
Currently the code assumes that positions are normalized to the range [0.0, 1.0) and represent absolute positions within a genome (i.e. a position times the length of the genome gives the absolute position).
However, simulation data could normalize the positions in other ways. I propose the following (better names for these are welcome):
-
absolute
- integer values giving absolute positions from 0, to the length of the genome minus 1 - absolute-normalized
- like
absolute` but re-normalized to the range [0, 1.0) by diving by the length of the genome -
relative
- integer values giving distances from the previous position -
relative-normalized
- likerelative
but re-normalized to the range [0, 1.0) by dividing by the length of the genome
When providing a simulation configuration, it should specify in which way the simulation data's positions are normalized. Further, when providing a training configuration, it should also be able to re-normalize the positions to any one of those options (or possibly others), as well as have a clearly specified default normalization.