nz_multirank

Load, rank, and sample a set of density n(z) realisations from a FITS file

File	number_density/nz_multirank/nz_multirank.py
Attribution	Juan P. Cordero
	Ian Harrison
URL

This module is designed to work with the number density part of the FITS files described in: http://github.com/joezuntz/2point/

Uncertainty in the redshift distributions is usually described by a nuisance parameter which allows the mean of the distribution to be shifted from a fiducial central value. This value is then marginalized in the pipeline. But higher order distribution moments which are not captured by this paramterization can propagate into the cosmological parameters and its uncertainty ignored.

An empirical approach to solve this is to provide multiple realisations of the redshift distributions n(z) containing realistic samples of the higher order moments. as well as the small redshift scale variance. We can then sample from them directly rather than using nuisance to capture the full effect of the shapes of the redshift distributions, as well as their internal correlation.

This module extends the load_nz_fits module to read multiple realisations, one per extension. It then ranks the realisations depending on the selected mode and maps it to a continuous hyper-parameter which can be sampled in the pipeline. The ranking is intended to provide a meaningful metric in the n(z) space, allowing for better sampling efficiency over random sampling of the realisations.

Assumptions

• Realisations are provided in FITS extensions NZ {NAME}_realisation_{NUMBER} starting with NUMBER=0, in correlated order and without skips

Setup Parameters

Name	Type	Default	Description
mode	str	mean	Ranking mode, use to define how the realisations are mapped to the hyperparameter. It has to be one of the following: mean, invchi, external
nz_file	str		Absolute or relative path to an n(z) file
data_set	str		Names of the extensions prefixes in the FITS files to load and save to the block
upsampling	int	1	The number of sample points output for each one in the file. n(z) is assumed flat between them. See notes above.
saved_stats	str		Numpy npy file from which to load rankings
bin_ranks	int 1d		The choice of which tomographic bins to generate statistics for
dimensions	int	2	The number of tomographic bins, and thus the dimensionality of map to generate
resume	bool	False	Whether to load a pre-computed map from parameters to rank
resume_map	str		Filename to load pre-computed map from

Input values

Section	Name	Type	Default	Description
ranks	rank_hyperparm_i	real		Hyperparameter mapped to a redshift distribution. If mode is separate, then i = 1…n_bins

Output values

Output values

Section	Name	Type	Description
wl_number_density	nz	int	Number of redshift samples
	nbin	int	Number of bins
	z	real 1d	Redshift sample values
	bin_	real 1d	n(z) at redshift sample values. bin_1, bin_2, …