Cloud and GPU-accelerated probabilistic modeling for national- and global-level datasets. This repository shows how to conduct inference and prediction using large datasets and rich graphical models via GPU-backed Markov chain Monte Carlo in NumPyro.
To get started, you'll need to make sure you have Jax enabled for CUDA and a CUDA-friendly NumPyro installation. Follow the Jax instructions here before installing NumPyro. Then, install the other dependencies using pip install -r requirements.txt.
Notebooks 00 and 01 include code and logic for preprocessing data associated with the brvehins car insurance dataset of 2.6M records from Brazil in 2011 and building a spatial statistical model for that data. This dataset was reproduced from the CASdatasets package developed for R and hosted on GitHub at this link. You can find more documentation on the origin of this data here. The file data/brvehins/brvehins_raw.parquet is a copy of the four RData files from the brvehins2 dataset listed on the linked Git repository.