murefi.core

class murefi.core.ParameterMapping(mapping: DataFrame, *, bounds: dict, guesses: dict)

Bases: object

Attributes:

bounds: (lower, upper) tuples for all parameters
coords: Groups the unique parameter ids by the kind of parameter.
guesses: Initial guesses for all parameters
mapping: Dictionary of parameter names or values (floats and strings) for each replicate
ndim: Dimensionality of the parameterization
order: Names of the model parameters
parameters: Maps unique parameters to the names of the corresponding model parameters.
theta_names: Names of unique parameters in the mapping

Methods

`as_dataframe`()	Re-creates the DataFrame representation of this parameter mapping.
`merge_vectors`(parameter_vectors)	Creates a full length (ndim,) parameter vector from a dictionary of parameter vectors.
`repmap`(theta_full)	Remaps a full parameter vector to a dictionary of replicate-wise parameters.

as_dataframe() → DataFrame

Re-creates the DataFrame representation of this parameter mapping.

It is NOT the identical DataFrame object it was initialized from!

property coords: Dict[str, Tuple[str]]

Groups the unique parameter ids by the kind of parameter.

Keys are in the form f”{pkind}_dim” to avoid conflicting with random variable names (see https://github.com/arviz-devs/arviz/issues/1642).

This dictionary can be used with pymc.Model(coords=coords) to ease creation of vector-shaped priors.

property mapping: dict: Dictionary of parameter names or values (floats and strings) for each replicate

merge_vectors(parameter_vectors: Dict[str, Sequence]) → tuple

Creates a full length (ndim,) parameter vector from a dictionary of parameter vectors.

Returns:

full_vectuple: the elements from the input parameter vectors in the order specified by .parameters

property parameters: OrderedDict: Maps unique parameters to the names of the corresponding model parameters.

repmap(theta_full: Sequence | dict) → Dict[str, Sequence]

Remaps a full parameter vector to a dictionary of replicate-wise parameters.

Args:

theta_full (array-like, dict): full parameter dict, vector or matrix

when dict:: keys are the unique parameter names (see ParameterMapping.parameters) values are float or Tensors, or numpy.ndarray (N_parametersets,)
when vector:: (N_parameters,) tuple, list, Tensor or numpy.ndarray with elements being scalar (float/Tensor)
when matrix:: (N_parameters, N_parametersets) numpy.ndarray

Returns:

theta_dict (dict): dictionary of replicate-wise parameter vectors/matrices

property theta_names: Tuple[str]: Names of unique parameters in the mapping