Contributed talk at Bayes Comp 2023
Abstract: Bayesian modeling in practice is an iterative process, in which a practitioner implicitly or explicitly follows the Bayesian workflow (Gelman et al 2020) to build models and inferences that are closest to the “reality” within the computational constraints. A composable model building capability is often desired as it makes developing bigger and more complex Bayesian models easier: for example, changing the priors of a collection of random variables. Moreover, a composable approach could enable more flexibility in constructing inferences that optimize for local model structure, thus have the opportunity to improve inference quality compared to using a general inference methods statistical packages offer (e.g., NUTS with different schemes of adaptation). In this talk, I will explain how adopting a functional programming perspective benefits the development of composable Bayesian modeling and programmable inference, with example using TensorFlow Probability on JAX (for the modeling part) and Blackjax (for the inference part).
conda create -n bayescomp23 python=3.10
conda activate bayescomp23
pip install -r requirements.txt-
golf_putting.ipynb: a notebook that demonstrates the iterative process of model building in a Bayesian workflow, with a functional programming princple to achieve composablity.
-
sparse_regression.ipynb: a simulation study with a sparse regression model, with a similar functional programming approch.
-
blackjax_deepdive.ipynb: a notebook that demonstrates the use of low level Blackjax to diagnose the performance of sampling routine.