Vanilla implementation of Gaussian Process in PyMC3?

I see. I didn’t modify the distances data, but I wonder if it’s either because:

  1. I’m using the covariance matrix instead of the Cholesky decomposition, or
  2. I’m not calculating the covariance matrix correctly (I’m doing it manually, based on what I saw from the PyMC source code… It’s highly likely that I made some mistake while trying to interpret it :sweat_smile:).

Could you show me how you’re then using the decomposition to build the GP?