Prior

class uvex_transients.models.core.priors.Prior

Abstract base class for one-dimensional statistical priors.

A prior represents a probability distribution over real-valued numerical coordinates. The primary purpose of a prior is to generate random samples through the sample() method.

Subclasses should be implemented as frozen dataclasses and are responsible for validating their own parameters (_validate()) and providing the distribution’s log-density (_logpdf()). Everything else – sampling, pdf(), cdf(), logpdf(), logcdf() – is derived from _logpdf automatically.

See also

scipy.stats.sampling.NumericalInversePolynomial

Backs the generic _sample() fallback.

Notes

A minimal implementation looks like

@dataclass(frozen=True)
class NormalPrior(Prior):
    mean: float
    sigma: float

    def _validate(self):
        if self.sigma <= 0:
            raise ValueError(
                "`sigma` must be positive."
            )

    def _logpdf(self, x):
        return scipy.stats.norm.logpdf(
            x, loc=self.mean, scale=self.sigma
        )

This is already enough for sample() to work, via numerical inversion of the CDF built from _logpdf. If a fast closed-form or scipy.stats sampler is available, override _sample() too:

def _sample(self, rng, size):
    return rng.normal(
        self.mean, self.sigma, size=size
    )

Methods

cdf(x)	Evaluate the cumulative distribution function.
logcdf(x)	Evaluate the log cumulative distribution function.
logpdf(x)	Evaluate the log probability density function.
logpmf(x)	Evaluate the log probability mass function.
pdf(x)	Evaluate the probability density function.
pmf(x)	Evaluate the probability mass function.
registry()	dict[str, type[Prior]]: A copy of every concrete Prior subclass, keyed by DISTRIBUTION_NAME.
sample([size, rng])	Draw random samples from the prior.

Attributes

DISTRIBUTION_NAME	The public-facing name of this distribution prior class.
name	Human-readable name of the distribution.
support	The (lower, upper) support of the distribution.