Thanks for that ( jessegrabowski and verderis).
For manually defined gradients, this would be for NUTS sampling, right? I managed to get it working using MCMC-Metropolis-Hastings, though noticeably slower.
I’ve not defined a gradient manually myself before in pymc. Would something along the lines of this suffice → Using a “black box” likelihood function (numpy) — PyMC example gallery ?
If so, a steer as to how for force this into the NUTS sampler would be ideal.
Thanks,
mmn 