It may be possible to efficiently sample from the distribution for the maximum spot radius, while applying a simple emergence/decay law that makes the spot grow and shrink. This might be possible if we create a little extra layer before we call rspot**2 in the mean model, and perhaps multiply rspot by a matrix of shape (len(times), 1) with the relative radius scale factor.
We should be careful to code this in such a way that it can be turned on and off easily, since we won't want to use spot evolution for every fit.
It may be possible to efficiently sample from the distribution for the maximum spot radius, while applying a simple emergence/decay law that makes the spot grow and shrink. This might be possible if we create a little extra layer before we call
rspot**2in the mean model, and perhaps multiply rspot by a matrix of shape (len(times), 1) with the relative radius scale factor.We should be careful to code this in such a way that it can be turned on and off easily, since we won't want to use spot evolution for every fit.