samreay / binningissinning Goto Github PK

This repository provides a simple validation that redshift binning of distances causes lost information in covariance matrices, which leads to inflated model uncertainty.

To do this, we first contruct a toy supernova model. No observational uncertainty, no selection effects, bias corrections, contamination, nothing.

We then add, using the same methodology as has been used in past literature, systematic uncertainty to the observational covariance matrix by incoporating the shift in distances between systematics.

By comparing the unbinned to the binned fitting, you will be able to see that any systematic which affects both redshift and any other parameter (which is, as far as I know, all of them), will thus overestimate the uncertainty.

Run main.py to run/load fits and generate plots. Chains are saved in the chains directory (delete this to refit), and plots are in the aptly named plots directory.

To install, ensure you have Python 3.6+ and dependencies (pip install -r requirements).

Here are the main two take away plots: