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.
Contract and evidence links¶
See Distribution kernels, the generated Scientific contract registry, and Finite power-law removable limit.
Canonical import example¶
from jaxstro.numerics.distributions import powerlaw_ppf