I have a very computationally expensive likelihood function. Using scipy’s maximum likelihood, I’m able to get a good estimate of my parameters. However, I’d like to get the distribution of those parameters and better statistics using NUTS. Is there a way to give NUTS my maximum likelihood estimates as a place to start? I think this would speed up NUTS significantly for the purposes of what I’m attempting to do given my likelihood function’s high computational cost. I’d rather not use MH-MCMC if possible. If NUTS doesn’t work, what’re other MCMC algorithms available through PYMC3 that allows me to use my maximum likelihood optimization outputs?

Thanks so much,

Zach