dais

Repository with code used for the paper "Doubly Adaptive Importance Sampling" by Willem van den Boom, Andrea Cremaschi and Alexandre H. Thiery (2024, arXiv:2404.18556).

Description of files

• dais/ is a Python module that provides an implementation of doubly adaptive importance sampling (DAIS) as well as other methods used for comparison. Other Python scripts import it.

• notebooks/control_variate.ipynb contains code for Figure 1 on the advantage of using Stein's identity.

• notebooks/banana_mixture.ipynb produces the figures for the two-dimensional synthetic examples (Figure 2, S2 and S3).

• notebooks/regression.ipynb produces the figures for the logistic regression examples (Figure 3, S4 to S7, and S9). It reads in data from data_regression/. It takes a bit longer to run, especially when the amount of RAM is less than 16GB.

• notebooks/convergence_banana.ipynb and [notebooks/convergence_gaussian.ipynb] (notebooks/convergence_gaussian.ipynb) contain code on monitoring of convergence (Figure S1).

• environment.yml can be used to create a conda environment similar to the one with which the code was tested on Ubuntu. Also, a suitable conda environment can be created as follows:

conda create -n dais -c conda-forge blackjax jupyter pandas papermill rpy2

Then, the results in the paper can be reproduced by running

conda activate dais
cd notebooks/
papermill control_variate.ipynb control_variate_output.ipynb --log-output
papermill banana_mixture.ipynb banana_mixture_output.ipynb --log-output
papermill regression.ipynb regression_output.ipynb --log-output
papermill convergence_banana.ipynb convergence_banana_output.ipynb --log-output
papermill convergence_gaussian.ipynb convergence_gaussian_output.ipynb --log-output

The code for the synthetic inverse problem (Figure S8) is available from the GitHub repository for arXiv:2404.18556v1.