Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Probability distributions

Owner import path

jaxstro.numerics.distributions

Purpose

Log-density, CDF, and inverse-CDF kernels for normal, lognormal, finite power-law, and truncated-normal families.

Public records and callables

normal_logpdf, normal_cdf, normal_ppf, lognormal_logpdf, lognormal_cdf, lognormal_ppf, powerlaw_logpdf, powerlaw_cdf, powerlaw_ppf, truncated_normal_logpdf, truncated_normal_cdf, and truncated_normal_ppf.

Shape and dtype expectations

Parameters and evaluation points are broadcast-compatible floating arrays. Inverse CDF inputs lie on the documented unit-interval domain.

JAX transforms and AD classification

Kernels compose with JIT, VMAP, and smooth-pathwise AD on regular domains. The finite power-law family uses a smooth removable-singularity formulation through alpha=-1.

Failure behavior

Support policy is explicit: log densities may return negative infinity, CDFs may saturate, and invalid quantile inputs are outside the contract.

See Distribution kernels, the generated Scientific contract registry, and Finite power-law removable limit.

Canonical import example

from jaxstro.numerics.distributions import powerlaw_ppf