As we start working on combined probes we will need to keep track of complicated combinations of two-point functions. So it is useful to put together a well-defined file format to avoid ambiguity when describing a set of observed or simulated quantities.
This repository describes a specification for such a file format and will collect code to process it.
See the initial notes file for some early thoughts that led to this format, and the included fits file as an example.
A 2-point data file is a FITS file containing extension headers with specified contents.
It should contain:
- one or more 2PTDATA extensions, each containing a two-point function between two specified types of quantity
- zero or more NZDATA extensions, each containing a collection of redshift distributions
- zero or more COVDATA extensions, each containing a covariance matrix for all the data in the file
- zero or more WINDATA extensions, each containing a collection of window functions of angle
A two point measurement is a correlation function of two quantities, usually as a function of some angular measurement (or angular wavenumber in Fourier space). Many different quantities can be used in two-point measurements. For example, if you see the quantity C^EE_ell, that could be a CMB polarization measurement or a galaxy shear measurement.
Three-letter quantity codes are used in the format described here to define more carefully what quantities are being used. The First letter describes the type of object or field being measured, the second what kind of statistical measurement is being used, and the third the basis (real, Fourier, or any other).
For example, the code GPF would represent Galaxy position C_ell values: the G stands for galaxy, the P for position, and the F for Fourier. A collection of other values are defined here, but more will be added as we progress. Not all possible combinations are valid.
First letter:
G : Galaxy
C : CMB
L : Clusters
P : Point-spread function
Second letter:
E : E-mode, for polarization (for C) or shear (for G)
B : B-mode
K : convergence
+ : For xi_+(theta)
- : For xi_-(theta)
Third letter:
F : Fourier space, such as C_ell
R : For real space such as correlation functions
C : For COSEBIS (?)
A 2PTDATA Binary Table extension is intended to contain the actual 2-point data values, for any required angular bin combinations, for a specific pair of quantities, such as galaxy shear E-mode, B-mode, position, xi(theta) CMB convergence, or any other quantities that can be correlated as a function of angle or angular scale.
The extension must contain the following parameters in its header:
2PTDATA = T // This sentinel marks the extension as 2PTDATA
QUANT1 = GPF // See "quantity codes" above
QUANT2 = GPF
NANGLE = 20 // The number of angular bins.
NBIN_1 = 3 // The number of redshift/distance bins.
NBIN_2 = 3 // The number of redshift/distance bins.
WINDOWS = SAMPLE // The form of the window function - at the moment this must be SAMPLE.
SIMULATD = T // This is simulated data
BLINDED = F // This is blinded data
For many 2-point data sets (but not, for example, CMB ones), the extension must also contain these header parameters:
KERNEL_1 = NZ_SOURCE // The n(z) data set for quantity 1. This is a reference to an NZDATA file elsewhere in the fits file
KERNEL_2 = NZ_SOURCE
The extension must then contain these data columns:
BIN1 8-byte integer // Integer, the index of the first bin
BIN2 8-byte integer // Integer, the index of the second bin
ANGBIN 8-byte integer // The angular bin index
VALUE 8-byte real // The data value for this combination.
Optionally, it can also contain this column:
ANG 8-byte real
An NZDATA Binary Table extension is intended to contain the number density distributions as a function of redshift for a quantity.
Not that we use the term "bin" to mean two things here - tomographic bins, meaning selections of galaxies hopefully in a particular redshift range, and histogram bins, the small division in redshift that we get n(z) measured in.
The extension must contain the following parameters in its header:
NZDATA = T // This sentinel marks the extension as n(z) data
EXTNAME = NAME // The name of this n(z) kernel.
NBIN = 5 // Integer number of tomographic bins
NZ = 100 // Integer number of histogram bins
The extension must then contain these data columns:
Z_LOW 8-byte real // Real, the z value for the lower end of each redshift histogram bin
Z_MID 8-byte real // Real, the z value for a middle point of each redshift histogram bin
Z_HIGH 8-byte real // Real, the z value for the upper end of each redshift histogram bin
BIN1 8-byte real // Real, the n(z) value for this histogram bin for the first tomographic bin
etc. BIN2, BIN3, etc.
A covdata extension contains a covariance matrix for all the data points in the file, stored as an image HDU.
Although the ordering of the points can be derived from the order of the data in the file, for convenience the header also includes information about which chunks of the matrix correspond to which data.
The header must contain these keys:
COVDATA = True // Sentinel for covariance matrix
EXTNAME = COVMAT // Name of this covariance matrix
NAME_0 = 'SHEAR_SHEAR' // Name of the 2PTDAT extension of the first chunk of covmat
STRT_0 = 0 // Integer, starting position of the first data set stored here
LEN_0 = 300 // Integer, length of the first data set stored here
NAME_1 = 'GGL' // Name of the 2PTDAT extension of the second chunk of covmat
STRT_1 = 300 // Integer, starting position of the second data set stored here
LEN_1 = 180 // Integer, length of the second data set stored here
The image in the extension should be of type 8-byte reals and be a square representation of the covariance matrix. The order of both the chunks of data and the values within each chunk must exactly match the order of the data in the file.
We have not yet considered this extension properly. At the moment only WINDOWS=sample is accepted.
This function will plot the correlation functions like xi_+, xi_-, gammat_t, w(theta) and the redshift distributions. You can use it by loading the two point file into the TwoPointFile class and call the plotting function in the following way:
T1 = twopoint.TwoPointFile.from_fits(filename1 + '.fits')
T1.plots(filename1)
The plots will be saved with the root name filename1. Options for a single file:
T1.plots(filename1, colormap='viridis', savepdf=False, latex=True, plot_spectrum=True, plot_kernel=True, plot_cov=True, cov_vmin=None, sharey=True, save_pickle=False, load_pickle=False, remove_pickle=True,label_legend ='', blind_yaxis=False, shade_until=None)
These are the default options. You can change the colormap, choose to also save pdf's besides png's, turn off the latex font and choose to not plot the spectrum, the kernel or the covariance. If save_pickle is true, it will save a pickle file so that you can tweak your plot afterwards loading this file. The y-axis can also be blinded by setting blind_yaxis to True. Here are some example plots that this function produces:
You can also make a plot comparing two twopoint files. This is only enabled for the spectra at the moment, but can be done for kernels too if there is interest. Here there is an example of how to do it:
T1 = twopoint.TwoPointFile.from_fits(filename1 + '.fits')
T2 = twopoint.TwoPointFile.from_fits(filename2 + '.fits')
Call the plots function twice, one for each file, with the same root name. The first time you call it, set save_pickle to True and the second time, set load_pickle to True. The first time it will save a pickle file that will allow to continue the plot the second time you call it, loading the same pickle file.
T1.plots(filename1+filename2, plot_kernel = False, plot_cov = False, save_pickle = True, label_legend = 'Filename 1 label')
T2.plots(filename1+filename2, plot_kernel = False, plot_cov = False, load_pickle = True, label_legend = 'Filename 2 label')