nnpd.nre.distributions¶

Distribution interfaces, independent bounded priors, and exact samplers.

Classes¶

`Parameter`
`Simulator`	Helper class that provides a standard way to create an ABC using
`Prior`	Helper class that provides a standard way to create an ABC using
`IndependentPrior`	Uniform, truncated normal, bounded signed exponential, or grid per axis.

Functions¶

grid_axis(→ numpy.ndarray)

Module Contents¶

class nnpd.nre.distributions.Parameter¶

name: str¶

low: float¶

high: float¶

options: dict¶

nnpd.nre.distributions.grid_axis(low: float, high: float, step: float) → numpy.ndarray¶

class nnpd.nre.distributions.Simulator¶

Bases: abc.ABC

Helper class that provides a standard way to create an ABC using inheritance.

parameters: tuple[Parameter, ...]¶

observation_dim: int¶

abstractmethod sample(theta: numpy.ndarray, rng: numpy.random.Generator) → numpy.ndarray¶: One observation per parameter row; result shape (…, observation_dim).

abstractmethod log_likelihood(x: numpy.ndarray, theta: numpy.ndarray) → numpy.ndarray¶: Optional exact reference, not used by the NRE training objective.

class nnpd.nre.distributions.Prior¶

Bases: abc.ABC

Helper class that provides a standard way to create an ABC using inheritance.

parameters: tuple[Parameter, ...]¶

abstractmethod sample(size: int, rng: numpy.random.Generator) → numpy.ndarray¶: Draw shape (size, number_of_inferred_parameters).

abstractmethod log_prob(theta: numpy.ndarray) → numpy.ndarray¶: Log density or log mass with respect to the declared support measure.

property support_axes: tuple[numpy.ndarray | None, ...]¶: None means continuous; an array means discrete counting measure.

property measure: str¶

class nnpd.nre.distributions.IndependentPrior(parameters: tuple[Parameter, ...], definition: dict)¶

Bases: Prior

Uniform, truncated normal, bounded signed exponential, or grid per axis.

A negative exponential rate is valid on a finite box; it is not an unbounded exponential distribution. Grid log_prob returns a mass, never a density. ‘by_parameter’ may override the default family for particular coordinates.

families¶

parameters¶

kinds¶

property support_axes¶: None means continuous; an array means discrete counting measure.

sample(size: int, rng: numpy.random.Generator) → numpy.ndarray¶: Draw shape (size, number_of_inferred_parameters).

log_prob(theta: numpy.ndarray) → numpy.ndarray¶: Log density or log mass with respect to the declared support measure.