Reference / binference.infer

binference.infer

binference.infer(likelihoods, prior=None, *, bins=None, selection=None, selection_mode=None, method='auto', n_samples=None, rng=None, progress=False, check=True, **engine_kwargs)

Infer a population distribution from per-object likelihoods in one call.

Parameters:

likelihoodsLikelihoods, PopulationModel, or array_like (N, K)

Per-object likelihoods over bins (see Likelihoods for converting posteriors, samples, Gaussian measurements, response matrices or classifier outputs). A raw array is treated as likelihoods over categorical bins (or bins).

priorPopulationPrior, optional

Population prior (default Dirichlet(1.0)). Use GammaRates to infer rates.

binsBins, optional

Bins for a raw array input.

selectionarray_like (K,), callable or Selection, optional

Detection probability (fractions) or exposure (rates) per bin.

selection_mode{“on_measurement”, “on_truth”}, optional

Required with an array or callable selection (see Selection).

method{“auto”, “nuts”, “gibbs”, “map”, “em”, “laplace”, “vb”, “bootstrap”}

Inference engine. "auto" (default) picks Gibbs, Laplace or NUTS from rules set by a benchmark of the engines (Auto); its options target_ess, gibbs_max_missing, max_gibbs_draws and nnz_limit go in engine_kwargs, and the choice and its reason are in result.info["auto_method"] and result.info["auto_reason"]. "nuts" samples the exact posterior with any prior; "em" is MAP by EM on the simplex (algorithm="em") rather than mix-SQP; it needs a conjugate prior, and a jointly inferred selection or response runs L-BFGS.

n_samplesint, optional

Draws per chain (NUTS, Gibbs, and "auto") or in total (Laplace, VB, bootstrap). Under "auto" it fixes the run length (see Auto).

rngint, numpy.random.Generator, or None

Random number generator or seed (SPEC 7).

progressbool

Show progress bars.

checkbool

Warn about convergence problems after sampling (NUTS, Gibbs).

**engine_kwargs

Passed to the engine (e.g. n_chains, n_warmup, target_accept for NUTS; see binference.samplers).

Returns:

PopulationResult or PointEstimate

Posterior draws, or the MAP estimate for method="map" and "em".

Examples

>>> import numpy as np, binference
>>> rng = np.random.default_rng(1)
>>> bins = binference.Bins.linear(0.0, 1.0, 5)
>>> z = rng.uniform(0, 1, 300)
>>> lk = binference.Likelihoods.from_gaussian(z + rng.normal(0, 0.05, 300), 0.05, bins)
>>> res = binference.infer(lk, rng=0)
>>> res.mean.shape
(5,)