We describe a procedure for modelling strong lensing galaxy clusters with parametric methods, and to rank models quantitatively using the Bayesian evidence. We use a publicly available Markov chain Monte-Carlo (MCMC) sampler ('Bayesys'), allowing us to avoid local minima in the likelihood functions. To illustrate the power of the MCMC technique, we simulate three clusters of galaxies, each composed of a cluster-scale halo and a set of perturbing galaxy-scale subhalos. We ray-trace three light beams through each model to produce a catalogue of multiple images, and then use the MCMC sampler to recover the model parameters in the three different lensing configurations. We find that, for typical Hubble Space Telescope (HST)-quality imaging data, the total mass in the Einstein radius is recovered with ~1-5% error according to the considered lensing configuration. However, we find that the mass of the galaxies is strongly degenerated with the cluster mass when no multiple images appear in the cluster centre. The mass of the galaxies is generally recovered with a 20% error, largely due to the poorly constrained cut-off radius. Finally, we describe how to rank models quantitatively using the Bayesian evidence. We confirm the ability of strong lensing to constrain the mass profile in the central region of galaxy clusters in this way. Ultimately, such a method applied to strong lensing clusters with a very large number of multiple images may provide unique geometrical constraints on cosmology.

emcee is an extensible, pure-Python implementation of Goodman & Weare's Affine Invariant Markov chain Monte Carlo (MCMC) Ensemble sampler. It's designed for Bayesian parameter estimation. The algorithm behind emcee has several advantages over traditional MCMC sampling methods and has excellent performance as measured by the autocorrelation time (or function calls per independent sample). One advantage of the algorithm is that it requires hand-tuning of only 1 or 2 parameters compared to $sim N^2$ for a traditional algorithm in an N-dimensional parameter space. Exploiting the parallelism of the ensemble method, emcee permits any user to take advantage of multiple CPU cores without extra effort.

Pangloss reconstructs all the mass within a light cone through the Universe. Understanding complex mass distributions like this is important for accurate time delay lens cosmography, and also for accurate lens magnification estimation. It aspires to use all available data in an attempt to make the best of all mass maps.

HumVI creates a composite color image from sets of input FITS files, following the Lupton et al (2004, ascl:1511.013) composition algorithm. Written in Python, it takes three FITS files as input and returns a color composite, color-saturated png image with an arcsinh stretch. HumVI reads the zero points out of the FITS headers and uses them to put all the images on the same flux scale; photometrically calibrated images produce the best results.

XDGMM uses Gaussian mixtures to do density estimation of noisy, heterogenous, and incomplete data using extreme deconvolution (XD) algorithms which is compatible with the scikit-learn machine learning methods. It implements both the astroML and Bovy et al. (2011) algorithms, and extends the BaseEstimator class from scikit-learn so that cross-validation methods work. It allows the user to produce a conditioned model if values of some parameters are known.

empiriciSN generates realistic supernova parameters given photometric observations of a potential host galaxy, based entirely on empirical correlations measured from supernova datasets. It is intended to be used to improve supernova simulation for DES and LSST. It is extendable such that additional datasets may be added in the future to improve the fitting algorithm or so that additional light curve parameters or supernova types may be fit.

qp manipulates parametrizations of 1-dimensional probability distribution functions, as suitable for photo-z PDF compression. The code helps determine a parameterization for storing a catalog of photo-z PDFs that balances the available storage resources against the accuracy of the photo-z PDFs and science products reconstructed from the stored parameters.

paltas conducts simulation-based inference on strong gravitational lensing images. It builds on lenstronomy (ascl:1804.012) to create large datasets of strong lensing images with realistic low-mass halos, Hubble Space Telescope (HST) observational effects, and galaxy light from HST's COSMOS field. paltas also includes the capability to easily train neural posterior estimators of the parameters of the lensing system and to run hierarchical inference on test populations.

ovejero conducts hierarchical inference of strongly-lensed systems with Bayesian neural networks. It requires lenstronomy (ascl:1804.012) and fastell (ascl:9910.003) to run lens models with elliptical mass distributions. The code trains Bayesian Neural Networks (BNNs) to predict posteriors on strong gravitational lensing images and can integrate with forward modeling tools in lenstronomy to allow comparison between BNN outputs and more traditional methods. ovejero also provides hierarchical inference tools to generate population parameter estimates and unbiased posteriors on independent test sets.

baobab generates images of strongly-lensed systems, given some configurable prior distributions over the parameters of the lens and light profiles as well as configurable assumptions about the instrument and observation conditions. Wrapped around lenstronomy (ascl:1804.012), baobab supports prior distributions ranging from artificially simple to empirical. A major use case for baobab is the generation of training and test sets for hierarchical inference using Bayesian neural networks (BNNs); the code can generate the training and test sets using different priors.