binference

Population and hierarchical inference with PDFs.

Many objects are each measured with noise: galaxies, merger events, patients, classified documents. binference reads each object's likelihood, posterior, samples, binned measurement or classifier output. From these it infers the population the objects come from, with calibrated uncertainty, and each object's posterior under that population.

The population inferred from 2000 simulated objects, under the default Dirichlet prior (top) and a Gaussian-process prior (bottom). Each panel shows the posterior median, its 68% and 95% intervals, the true density (dashed) and the stacked estimate (dotted).

Install

pip install "binference[plot]"

The installation page lists the extras and the supported Python versions.

Example

import binference

bins = binference.Bins.linear(0.0, 3.0, 15)

# a simulated stand-in for your arrays: measured values and their Gaussian errors
mock = binference.simulate.mock_1d(2000, bins, rng=42)
z_obs, z_err = mock.measurements["mean"], mock.measurements["sigma"]

lk = binference.Likelihoods.from_gaussian(z_obs, z_err, bins)
result = binference.infer(lk, rng=1)  # method="auto" chooses the engine
print(result.summary())

Documentation

• Quick start
• Is binference right for my data?
• Tutorials
• Full documentation

To contribute, see CONTRIBUTING.md.

Status and citing

binference is alpha software, and its API may change before 1.0. If you use it, please cite it as described on the citing page.

License

MIT; see LICENSE.