Guides / Prepare your inputs

Prepare your inputs¶

binference needs one thing from your data. For each object \(i\) and bin \(k\) it needs the bin-averaged likelihood \(L_{ik}\). That is how probable the object’s data are if its true value lies in bin \(k\). A Likelihoods object holds these values with their Bins. Its constructors turn the usual data products into likelihoods.

Find your input¶

What you have for each object	Constructor	Page
a measurement and its Gaussian error	`Likelihoods.from_gaussian`	Gaussian measurements
a likelihood function, or a SciPy distribution	`Likelihoods.from_density`	Likelihood functions and distributions
likelihood values per bin	`Likelihoods(L, bins)`, `from_table`, `from_log`	Likelihood tables
a posterior per bin, computed under some prior	`Likelihoods.from_table(..., interim=)`	Posteriors under an interim prior
samples from a posterior or a likelihood	`Likelihoods.from_samples`	Posterior samples
a measured bin, or a classifier’s score or label	`Likelihoods.from_binned`	Binned measurements and classifiers

If your code reports posteriors, they already contain a prior. Say which prior. Otherwise the fit biases the population toward that prior (Posteriors under an interim prior).

Bins¶

Every constructor takes the bins of the true value. Build them first:

import numpy as np
import binference

bins = binference.Bins.linear(0.0, 3.0, 30)              # 30 equal bins on [0, 3]
logbins = binference.Bins.log(1e-2, 1e2, 20)             # equal in log z
irregular = binference.Bins([0.0, 0.2, 0.5, 1.0, 3.0])   # explicit edges
grid = binference.Bins.grid(np.linspace(-0.5, 2.0, 26), np.linspace(-5.0, 15.0, 41))
classes = binference.Bins.categories(["star", "galaxy", "quasar"])
print(bins.K, grid.shape, grid.K, classes.labels)

Bins know their edges, centers and volumes (widths in 1-D), and index returns the bin of each value. A 2-D grid is flattened in C order: unflatten and flatten move between the vector of \(K\) values and the grid. Categories have unit volumes and no geometry. Choose bins says how many bins to use and where to put them.