Deep inference of simulated strong lenses in ground-based surveys

We present data used in the paper "Deep inference of simulated strong lenses in ground-based surveys", published in JCAP. The large number of strong lenses discoverable in future astronomical surveys will likely enhance the value of strong gravitational lensing as a cosmic probe of dark energy and dark matter. However, leveraging the increased statistical power of such large samples will require further development of automated lens modeling techniques. We show that deep learning and simulation-based inference (SBI) methods produce informative and reliable estimates of parameter posteriors for strong lensing systems in ground-based surveys. We present the examination and comparison of two approaches to lens parameter estimation for strong galaxy-galaxy lenses -- Neural Posterior Estimation (NPE) and Bayesian Neural Networks (BNNs). We perform inference on 1-, 5-, and 12-parameter lens models for ground-based imaging data that mimics the Dark Energy Survey (DES). We find that NPEs outperform BNNs, producing posterior distributions that are more accurate, precise, and well-calibrated for most parameters. For the 12-parameter NPE model, the calibration is consistently within <10% of optimal calibration for all parameters, while the BNN is rarely within 20% of optimal calibration for any of the parameters. Similarly, residuals for most of the parameters are smaller (by up to an order of magnitude) with the NPE model than the BNN model. This work takes important steps in the systematic comparison of methods for different levels of model complexity. Training Datasets: Dataset used in our 1-parameter experiments: 1param_200k_train.pkl Entire 200k dataset is used as a training set for both NPE and BNN models. An additional 50k BNN validation set can be generated using 1param_model_valid_50k.yaml. (In all experiments NPE training procedure internally generates a validation set by using 10% of the training set.) Dataset used in our 5-parameter experiments: 5param_model_training_500k_Aug29.pkl The first 400k images are used as a training set for both NPE and BNN experiments, while the last 100k images are used as a validation set for BNN. Dataset used in our 12-parameter experiments: 12_model_training_des_1M.pkl The first 800k images are used as a training set for both NPE and BNN experiments, while the last 200k images are used as a validation set for BNN. All test datasets can be generated using the following yaml files: 1 parameter test set - 1param_test_set.yaml 5 parameter test set - 5param_test_set.yaml 12 parameter test set - 12param_test_set.yaml 12 parameter source - 12param_test_set_source.yaml (only for plotting source images) The NPE and BNN models for each level of complexity are as follows: 1 parameter: Seed 42: 1param_seed42.pkl Seed 465: 1param_seed465.pkl Seed 839: 1param_seed839.pkl BNN: 1param_BNN.h5 5 parameter: Seed 42: 5param_seed42.pkl Seed 465: 5param_seed465.pkl Seed 839: 5param_seed839.pkl BNN: 5param_BNN.h5 12 parameter: Seed 42: 12param_seed42.pkl Seed 465: 12param_seed465.pkl Seed 839: 12param_seed839.pkl BNN: 12param_BNN.h5