Using pm.Data to predict on two inputs for sample_posterior_predictive; why is there no change in the results?

The sampler would start over. Therefore you should include both old and new observations. There is no simple solution to do incremental updating in mcmc, but there are some less than perfect options such as using kennel density estimation as described in here: https://docs.pymc.io/notebooks/updating_priors.html

If speed is not an issue it’s better to just fit the model with the larger dataset.

2 Likes