Feature rescaling needed for bayesian neural networks?

[HG1]

Hi Ferrine,
I did not quite understand this:

@ferrine

and first differences dor sequential data might be a good feature regardless Bayesian stuff.

Could you explain a bit?