elisa.infer.samplers.util#

class ModelInfo(ndim: int, init: dict[str, NDArray[floating]], init_ravel: NDArray[floating], unravel: Callable[[NDArray[floating]], dict[str, NDArray[floating]]], log_prob_fn: Callable[[dict[str, NDArray[floating]]], float], postprocess_fn: Callable[[dict[str, NDArray[floating]]], dict[str, NDArray[floating]]], params_names: list[str], params_dtype: list[tuple[str, dtype, tuple[int, ...]]], deterministic_names: list[str], deterministic_dtype: list[tuple[str, dtype, tuple[int, ...]]])[source]#

Bases: NamedTuple

Model information.

Methods

`count`(value, /)	Return number of occurrences of value.
`index`(value[, start, stop])	Return first index of value.

ndim: int#: Model dimension.

init: dict[str, NDArray[floating]]#: Initial parameters values in unconstrained space.

init_ravel: NDArray[floating]#: Raveled initial parameters values in unconstrained space.

unravel: Callable[[NDArray[floating]], dict[str, NDArray[floating]]]#: Function to unravel parameters values.

log_prob_fn: Callable[[dict[str, NDArray[floating]]], float]#: Log probability function given parameters in unconstrained space.

postprocess_fn: Callable[[dict[str, NDArray[floating]]], dict[str, NDArray[floating]]]#: Postprocess function given parameters in unconstrained space.

params_names: list[str]#: Names of parameters.

params_dtype: list[tuple[str, dtype, tuple[int, ...]]]#: NumPy dtypes of parameters in constrained space.

deterministic_names: list[str]#: Names of deterministic sites.

deterministic_dtype: list[tuple[str, dtype, tuple[int, ...]]]#: NumPy dtypes of deterministic sites.

get_model_info(model: Callable, init_strategy: Callable = <function init_to_uniform>, model_args: tuple = (), model_kwargs: dict | None = None, forward_mode_differentiation: bool = False, validate_grad: bool = True, rng_seed: int = 42) → ModelInfo[source]#: Get model information.

ravel_params_names(name: str, shape: tuple[int, ...]) → list[str][source]#: Ravel parameter names.

class UniformReparam[source]#

Bases: Reparam

Reparameterize a distribution to a Uniform over the unit hypercube.

Most univariate distribution uses inverse CDF for reparameterization.

Methods

__call__(name, fn, obs)

uniform_reparam_transform(d)[source]#: A helper for UniformReparam to get the transform that maps a uniform distribution over a unit hypercube to the target distribution d.

uniform_reparam_model(model: Callable, model_args: tuple = (), model_kwargs: dict | None = None, rng_seed: int = 42) → ModelInfo[source]#