Frequentist uncertainties for DRE with parameterized ensembles, based on https://arxiv.org/abs/2506.00113.
The experiments directory contains two directories, gaussian and qg, corresponding to the two case studies in the paper. The .py scripts generate results, and the .ipynb notebooks generate the figures from the results. The scripts take a long time to run on one GPU ($\mathcal{O}$(1 day) for the Gaussian case study, $\mathcal{O}$(1 week) for the quark/gluon case study), so if you are running additional small experiments you will likely want to either reduce the number of basis functions $M$, the number of trainings, or the number of trials per training. If your goal is to reproduce the results from the paper, or if you want to run larger experiments, you will likely need to parallelize over trainings; this is not difficult since the trainings are independent of each other, but the easiest way to implement this is environment specific, so we provide the serial version for generality.
The tutorials directory currently includes a step-by-step explanation of an implementation of $w_i f_i$ ensembles for a scaled-down version of the Gaussian example considered in the paper.
