Tutorials¶
Each tutorial is a complete analysis that runs from simulated data to a result. The groups follow what you have for each object. Start with the group that matches your data; the last group applies to any data.
The documentation build runs every tutorial, so their outputs are exactly what the code produces. Each runs in about a minute or less on a shared four-core container with one BLAS thread per process.
Gaussian measurements¶
A measured value and its error for each object.
Gaussian measurements: a redshift distribution: a redshift distribution, compared with stacking, with derived quantities and single galaxies.
Priors and smoothness: four priors on a sparse catalogue, and held-out objects to choose between them.
A two-dimensional population: a population on a grid of two measured quantities.
Empirical Bayes deconvolution: deconvolution, the NPMLE and shrinkage estimates for each unit.
Posteriors on a grid¶
Posteriors on a grid: a table of posteriors computed under an interim prior, and masses versus densities.
Posterior samples¶
Posterior samples: black-hole masses: black-hole masses from parameter-estimation samples and their prior.
Classifier outputs¶
Classifier outputs: class fractions under label shift: class fractions under label shift, from posteriors and the training prevalence, or from scores and a labelled set.
Binned counts and a response¶
Binned counts and a response: unfolding a spectrum: a spectrum unfolded from detector counts with a response matrix.
Any likelihood table¶
Mixed-stock analysis: stock proportions of a fishery’s catch from genotype likelihoods.
Incomplete catalogues, and checking an analysis¶
Selection effects and rates: occurrence rates with a known completeness, and a cut on the measured value.
Selection measured by injections: a mass distribution with a selection measured by injections.
Choosing an engine: every engine on one problem.
Validating an analysis: exact posteriors, simulation-based calibration and coverage.