Hey all! I am trying to use the multivariate normal random walk to forecast values in the future. The best way to do this (at least in JAGS) is to add nans to the observed data. I tried to do that using pymc3 because I want to try out some of the very nice VI tools, but I can’t quite get it to work…

Actually, random walk is modeling X_{t} - X_{t-1} \sim MvNormal(\theta), which makes missing value quite difficult to handle. I will need to think a bit more about it.

Multivariate Random Walk with missing values

Questions

junpenglao September 10, 2019, 10:00am 2

Did you have a look at the discussion in Multivariate Normal with missing inputs? I feel that the solution might apply.

Topic		Replies	Views
Handling missing values in predictor when outcome is a Multivariate Normal distribution v5	7	207	October 25, 2024
randomWalk predictions Questions	0	413	February 25, 2020
PyMC v5.14.0 introduces error in MultivariateNormal with NaN v5 development , bug , modeling	3	143	May 14, 2024
Multivariate Random Walk Regression v5	3	393	December 8, 2023
Multivariate Normal Imputation Error v5 aesara	1	540	December 19, 2022

Multivariate Random Walk with missing values

Related topics