Ah, I made a 1d entorhinal cortex a while ago. (Here) I’m curious how they compare!
Now, while it isn’t the same, the encoding in the entorhinal cortex shares some traits worth computer numbers. Powers of 2 are good for 1d (not others, and e is better. Here’s the related research article. ) but base 2 has a bit too many activations to be useful. It’d likely be better to use ~50 bits to represent 0-50, then ~50 bits to represent 0-2500, and so on. Though, it’s probably fine to keep lower than 50 if you’re using a spatial pooler.
For representing floating points, other than just representing 0-50^0 and 0-50^-1, I’d follow IEEE float or decimal formats, and just use scalar encoders for decimal and exponent value.