Consistent-Stochastic-Variational-Inference

This repository provides source code for the experiments in

Z. Xu and T. Campbell, The computatioal asymptotics of Gaussian variational inference.

Examples run and generate output using Python3.

• VI/ provides functions for inferences (CSVI/SVI and smoothed MAP)
• examples/ provides code to replicate examples and figures
• Datasets(both raw and processed) used in experiments are provided in examples/data/, including code that generates the synthetic datasets and processes real datasets

How to run the code

Each experiment should be run in its own folder (examples/synthetic_mixture/, examples/sparse_regression/, and examples/Gaussian_mixture/):

• first run ./run.sh to perform the experiment
• then run python3 plot.py to generate plots

Note: In examples/sparse_regression/, run python3 stan.py before running python3 plot.py---one of the plots uses samples from PyStan.