Github issue #2020 by 23pointsNorth : Hi, I have a model with a set of latent variables, which I am trying to evaluate. So far I have worked in a custom MCMC world and want to switch to HMC/NUTS. However, I was unable to have it as a drop in replacement, as the model variables in PyMC3 are Subtenso…

You can write a custom theano Op. The theano documentation contains a section about that, and you can have a look at the exoplanet example (which still isn’t a full blog post, but I’m working on :slight_smile:) https://gist.github.com/aseyboldt/57ebb66dd6597d6ea7a9d59f24f264e0 …

Getting observations of distribution variables whilst sampling

Questions

junpenglao June 15, 2017, 2:40pm 3

@23pointsNorth you can also find a similar discussion here:
https://github.com/pymc-devs/pymc3/issues/1925

Topic		Replies	Views
Defining custom multivariate Distribution with gradients Questions theano	5	1605	August 24, 2018
Using pm.DensityDist and customized likelihood with a black-box function Questions	10	6853	August 29, 2018
Getting the current sample variable value so it can be used to compute the observed inputs in a secondary model likelihood Questions	10	2267	June 25, 2018
Unobserved variables in blackbox likelihood Questions	11	1120	October 14, 2019
Defining a numeric (custom) likelihood function in PyMC3 Questions theano	2	3070	August 25, 2018

Getting observations of distribution variables whilst sampling

Related topics