PriorCVAE

Demonstration code of the publication "PriorCVAE: scalable MCMC parameter inference with Bayesian deep generative modelling"

Environment

Create the environment numpyro10_torch:

conda create -n numpyro10_torch python=3.8.15
conda activate numpyro10_torch
conda install -c conda-forge jax=0.3.25
conda install -c conda-forge numpyro=0.10.1
conda install pytorch=1.12.1 -c pytorch

conda install -c conda-forge matplotlib
conda install -c anaconda Jupyter
conda install -c conda-forge arviz
conda install -c conda-forge dill
conda install -c conda-forge mamba
mamba install -c conda-forge geopandas
conda install -c anaconda seaborn
mamba install -c conda-forge wandb

Google Colab

A runnable demo of the one-dimesiontal GP example comparing PriorVAE, PriorCVAE and GP inference with MCMC is available on Colab. Make sure to keep trained_models and mcmc folders in the root directory; trained_models contains pretrained neural networks (VAEs), and mcmc contains MCMC fits of models which are hard to run.

Update

Code in this repository uses PyTorch. It is recommended to use the JAX verion instead:

• JAX version of the PriorCVAE method: here