gwkokab.analysis.core.utils¶

Shared helpers for the analysis drivers.

PRNGKeyMixin gives an analysis a class-level PRNG key that splits on every access, so independent parts of a run get independent randomness from a single recorded seed. write_to_hdf5() and read_from_hdf5() are the two ends of the output format, and handle the encoding of non-array attributes – None, lists and dicts – into what HDF5 accepts. SampleTransformer is the hook by which a data loader changes the coordinates its events are expressed in, carrying the Jacobian along so densities stay normalised.

Classes¶

`IdentitySampleTransformer`	The transformer that changes nothing.
`PRNGKeyMixin`	Mixin class that provides a random number generator key and seed management for
`SampleTransformer`	Change of coordinates applied to per-event samples as they are read.

Functions¶

`from_structured`(→ tuple[numpy.ndarray, ...)	Converts a structured array to a 2D array.
`read_attrs_from_hdf5`(→ dict)	Reads attributes from a specified dataset in an HDF5 file.
`read_from_hdf5`(→ numpy.ndarray)	Reads data from an HDF5 file at the specified dataset path.
`to_structured`(→ numpy.ndarray)	Converts a 2D array to a structured array with given field names.
`write_to_hdf5`(→ None)	Writes data to an HDF5 file at the specified dataset path, with optional

Module Contents¶

class gwkokab.analysis.core.utils.IdentitySampleTransformer¶

Bases: SampleTransformer

The transformer that changes nothing.

Used when the data loader’s coordinates already match the model’s, so the samples pass through untouched and the log Jacobian determinant is zero.

log_abs_det_jacobian(samples: numpy.ndarray, transformed_samples: numpy.ndarray) → numpy.ndarray¶

The log Jacobian determinant of the identity, which is zero.

Parameters:

samples (np.ndarray) – The original samples.
transformed_samples (np.ndarray) – The transformed samples; identical to samples.

Returns:

Zeros of shape samples.shape[:-1].

Return type:

np.ndarray

transform(samples: numpy.ndarray) → numpy.ndarray¶

Return the samples unchanged.

Parameters:: samples (np.ndarray) – The samples to transform.
Returns:: samples, unchanged.
Return type:: np.ndarray

class gwkokab.analysis.core.utils.PRNGKeyMixin¶

Mixin class that provides a random number generator key and seed management for classes that require random number generation.

This mixin allows classes to initialize a random key based on a seed and provides a property to access the current random key, which is automatically split to ensure independent random streams across different parts of the code.

classmethod init_rng_seed(seed: int) → None¶

Initialise the class-level random key from a seed.

The key lives on the class rather than the instance, so every analysis object in a run draws from the same reproducible stream.

Parameters:: seed (int) – A non-negative integer seed.
Raises:: LoggedValueError – If seed is not an integer, or is negative.

property rng_key: jaxtyping.PRNGKeyArray¶

Accessor for the current random number generator key.

Each call to this property will split the key and return a new subkey, ensuring that different parts of the code can use independent random keys without interfering with each other.

Returns:: A new random key generated by splitting the current key.
Return type:: PRNGKeyArray

property seed: int¶

Accessor for the seed used to initialize the random number generator key.

Returns:: The seed used for RNG initialization.
Return type:: int

class gwkokab.analysis.core.utils.SampleTransformer¶

Bases: abc.ABC

Change of coordinates applied to per-event samples as they are read.

A data loader may express its events in coordinates the model does not use. Subclasses implement the forward map and its log Jacobian determinant, which is applied to the PE prior so that the reweighted density stays normalised across the change.