sheap.Profiles.Utils module

Profile Utilities

This module provides utility functions to support the definition, integration, and composition of spectral line and continuum profiles in sheap.

Functions

  • make_fused_profiles(funcs) :

    Combine multiple profile functions into a single callable that evaluates the sum of all components.

  • with_param_names(param_names) :

    Decorator to attach parameter names and count metadata to profile functions for consistent handling across the codebase.

  • make_integrator(profile_fn, method="broadcast") :

    Factory for JAX-based integrators of profile functions. Supports either broadcasting across batches or nested vmap evaluation.

  • build_grid_penalty(weights_idx, n_Z, n_age) :

    Construct a Laplacian smoothness penalty on 2D grids of host template weights (Z × age), useful for regularization.

  • trapz_jax(y, x) :

    Lightweight trapezoidal integration implemented with JAX.

Notes

  • All utilities are JAX-compatible and designed for use in differentiable fitting pipelines.

  • with_param_names ensures that each profile function exposes .param_names and .n_params attributes for downstream bookkeeping.

  • make_integrator can be used to integrate profiles per spectrum, per component, or across batches without manual looping.

build_grid_penalty(weights_idx, n_Z, n_age)[source]

Construct a Laplacian smoothness penalty over a 2D template grid.

Parameters:

  • weights_idx (list[int]) – Indices of weight parameters in the global parameter vector.

  • n_Z (int) – Number of metallicity bins.

  • n_age (int) – Number of age bins.

Returns:

penalty – Function (params) -> float computing the smoothness penalty.

Return type:

callable

make_fused_profiles(funcs)[source]

Fuse multiple profile functions into a single callable while preserving parameter metadata, including which parameters are linear.

make_integrator(profile_fn, method='broadcast')[source]

Create an integrator for profile functions. This works for 1D wavelength and 3D params n_sample,n_lines,n_params.

Parameters:

  • profile_fn (callable) – Profile function with signature (x, params) -> y.

  • method ({"broadcast", "vmap"}, optional) – Integration strategy: - “broadcast”: expand x for broadcasting across batches - “vmap”: use nested vectorization

Returns:

integrate – Function (x, params) -> integral returning integrated flux.

Return type:

callable

trapz_jax(y, x)[source]

JAX-compatible trapezoidal integration.

Parameters:

  • y (jnp.ndarray) – Function values.

  • x (jnp.ndarray) – Grid over which integration is performed.

Returns:

Approximate integral of y over x.

Return type:

jnp.ndarray

with_param_names(param_names, linear_param_names=None, linear_param_mask=None, profile_name=None)[source]

Decorate a profile function with parameter metadata.

Parameters:

  • param_names (iterable of str) – Ordered parameter names for the function.

  • linear_param_names (iterable of str, optional) – Names of parameters that enter linearly.

  • linear_param_mask (iterable of bool, optional) – Boolean mask of the same length as param_names indicating which parameters are linear.

Returns:

Decorated function with attributes:

  • param_names

  • n_params

  • linear_param_mask

  • linear_param_indices

  • nonlinear_param_indices

  • linear_param_names

  • nonlinear_param_names

Return type:

callable

Notes

Provide either linear_param_names or linear_param_mask. If neither is provided, all parameters are assumed nonlinear.

class GaussianSum(n, constraints=None, inequalities=None)[source]

Bases: object