After installing Docker, run the following commands in the root of the present repository.
docker build --tag 'dadvi_paper' .
Then you can run
docker run -it --platform linux/amd64 dadvi_paper /bin/bash
and, at the prompt, type
conda activate dadvi
This should put you in a conda environment with the dadvi package and its dependencies
successfully installed.
To install this repository, please first install pymc. The suggested way of doing this
(see here) is to run:
conda create -c conda-forge -n dadvi "pymc=5.6" bambi=0.13.0 python=3.9 "scipy<1.13" "numpy==1.25.2"
conda activate dadvi
Note that bambi is optional and only needed for some models. Also, the exact version dependencies
given here may not have to be followed exactly, but it's safest to keep them, as these are the ones
we tested.
You can then install the package using
pip install -e .
To run the tests, run the following command in the root of the repository:
python3 tests/run_tests.sh
To see how to use the package with PyMC, please take a look at the notebook "Radon example.ipynb" in the jupyter folder.
To run the experiments from our paper, "Black Box Variational Inference with a Deterministic Objective: Faster, More Accurate, and Even More Black Box", by Giordano, Ingram and Broderick, please look at github.com/martiningram/dadvi-experiments.