I’ve recently made the switch from PyMC3 to NumPyro for speed reasons, and am loving it so far. A project I was previously working on in PyMC3 required a custom prior distribution on a latent parameter. In PyMC3, I would have set up this model as follows:
import pymc3 as pm def bailerjones_lpdf(L): def lpdf(r): return 2*pm.math.log(r) - 3*pm.math.log(L) - (r/L) return lpdf L_ = 800 # in parsec with pm.Model() as model: r = pm.DensityDist('r', bailerjones_lpdf(L_), transform = pm.distributions.transforms.Log()) trace = pm.sample()
I’m at a loss on how get something like this to work in NumPyro, and can’t find any tutorials. Can anybody point me in the right direction?
Thanks so much!