Hi folks, below is a model where there is missing data in a MVN-modelled predictor matrix, and while I love how pymc automatically does missing value imputation, I notice that to achieve this a metropolis step is added to the sampling scheme, which prevents me from using things like nutpie and gpu a…

using x_Q triggers metropolis because QR gradients were only just implemented in pytensor 2.30.0 . PyMC just had a release last night to be compatible with pytensor 2.30, so if you update everything it should just work.

Missing data imputation for predictors without triggering metropolis?

Questions v5

mike-lawrence April 4, 2025, 1:56pm 3

Ha, what a wonderful coincidence. Thanks!

2 Likes

Topic		Replies	Views
Missing values in a model? Questions	12	4791	November 7, 2018
Missing values -- unexpected behavior Questions	0	465	February 19, 2020
Metropolis sampler do not sample (traceplot gives flat plots) Questions	11	2840	June 15, 2018
Multivariate Random Walk with missing values Questions	2	744	September 10, 2019
Sampling time is very long (minutes per sample) Questions	13	4561	November 2, 2017

Missing data imputation for predictors without triggering metropolis?

Related topics