This repository contains source code for Erickson et al. 2024 "Lens Modeling of STRIDES Strongly Lensed Quasars using Neural Posterior Estimation"
To reproduce all figures in the paper, use Paper/make_figures.ipynb. This will require a Zenodo download. To recreate figures, you do NOT need to download trained_models.tgz (26 GB!).
This code base heavily relies on the use of paltas (https://github.com/swagnercarena/paltas). Note that while updates from this project are being merged into the paltas main branch, the @smericks fork of paltas contains all necessary updates: https://github.com/smericks/paltas
See Notebooks for demonstrations of key analysis components:
- configurations for training neural networks (training_networks.ipynb)
- feeding images to a neural network to generate mass model posteriors (make_predictions.ipynb)
- hierarchical Bayesian inference for lens mass population models (hierarchical_inference.ipynb)
- hierarchical re-weighting of individual posteriors to account for out-of-distribution shift (posterior_reweighting.ipynb)
Note: some notebooks also require the Zenodo download