n(z) prior

A DESC Package to turn an ensemble of redshift distributions into a prior for photo-z uncertainty.

Usage

nz_prior can either be run on a QP ensemble or on a sacc file with QPnz tracers. Here is an example of the simplest use when a QP ensemble is provided:

import nz_prior as nzp
X = nzp.PriorX(q) 
#  X is the method to capture photometric uncertainty
#  q is the sacc file with QPnz tracers

Here is an example when a sacc file is provided instead:

X = nzp.PriorSacc(sacc_file, 
    model="Model",
    compute_crosscorrs="Full",
    zgrid=zgrid)

A couple of running options are important to note:

• model: the model used to capture the uncertainty in the n(z)'s within the sacc file.
  • "Shifts": the n(z)'s are shifted by a constant amount.
  • "ShiftsWidths": the n(z)'s are shifted by a constant amount and broadened by a constant width.
  • "PCA": the n(z)'s are represented as a linear combination of the principal components.
  • "Fourier": the n(z)'s are represented as a linear combination of Fourier modes.
  • "GP": the model assumes that the n(z)'s are a Gaussian Process.
  • "Comb": the n(z)'s are represented as a linear combination of evenly spaced Gaussians.
• compute_crosscorrs: whether to compute the cross-correlations between the n(z)'s or not.
  • "None": do not compute the cross-correlations.
  • "BinWise": compute the cross-correlations between the parameters associated with the same tomographic bin.
  • "Full": compute the cross-correlations between all the parameters across tomographic bins.
• zgrid: the zgrid on which to evaluiate the n(z)'s. If None, the zgrid of the sacc file is used.

Upon initializing the class, Gaussian priors for the model parameters will be calibrated. The prior can then be accessed as an attribute of the class:

mean, cov, chol = X.get_prior()

which returns the mean, covariance, and Cholesky decomposition of the prior.

Moreover, it is possible to sample from the prior:

samples = X.sample_prior()

which procudes a dictionary with the following structure Dict{param_names: value}. The keys are the names of the parameters and the values are the samples.

The samples can then be used to generate n(z)'s using the models in the nz_prior.models module.

Checkout the examples folder for more detailed use cases.

Installation

To install the package, run the following command in the root directory of the package:

pip install .

Note: make sure that you have installed sacc version 0.16 or higher.