Source code for trspecfit.spectra

"""
Spectrum generation functions for fitting.

This module provides the interface between trspecfit's model/component/parameter
(mcp) system and the fitting routines. It contains functions that generate
spectral data from model parameters during optimization.

The functions here are called by the fitting engine (fitlib.residual_fun) on
every iteration to compute the current model prediction, which is then compared
to experimental data.

Key Concepts
------------
- fit_model_mcp: Default spectrum generator using mcp.Model system
- Custom generators: Users can define alternative spectrum functions
  and specify them via Project.spec_fun_str
  ["x, par, plot_sum, args" is the typical fit function structure]

Architecture
------------
The fitting workflow is:
1. Optimizer proposes new parameter values
2. fitlib.residual_fun calls spectrum function (this module)
3. Spectrum function generates model prediction
4. Residual = data - model is computed and returned to optimizer
"""

from __future__ import annotations

from collections.abc import Sequence
from typing import Any, Literal, overload

import numpy as np

from trspecfit.eval_1d import evaluate_1d
from trspecfit.eval_2d import evaluate_2d
from trspecfit.graph_ir import ScheduledPlan1D, ScheduledPlan2D
from trspecfit.mcp import Model


#
# plot_sum=True always returns the summed spectrum; only plot_sum=False
# (1D component extraction) returns a list. Mirrored by the other
# fit_model_* overloads below.
@overload
def fit_model_mcp(
    x: Sequence[float] | np.ndarray,
    par: Sequence[float] | np.ndarray,
    plot_sum: Literal[True],
    model: Model,
    dim: int,
) -> np.ndarray: ...
@overload
def fit_model_mcp(
    x: Sequence[float] | np.ndarray,
    par: Sequence[float] | np.ndarray,
    plot_sum: bool,
    model: Model,
    dim: int,
) -> np.ndarray | list[np.ndarray]: ...
[docs] def fit_model_mcp( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, model: Model, dim: int, ) -> np.ndarray | list[np.ndarray]: """ Generate spectrum from mcp.Model for fitting or visualization. This is the default spectrum generation function used by trspecfit. It updates model parameters, evaluates the model, and returns either the complete spectrum or individual component spectra. Parameters ---------- x : array-like Independent variable axis (energy or time). Not directly used here as model contains its own axes, but required for fitting interface compatibility. par : list or array-like Parameter values in same order as model.parameter_names. These are the current values proposed by the optimizer during fitting. plot_sum : bool Component return mode: - True: Return sum of all components (used during fitting) - False: Return list of individual component spectra (for visualization) model : mcp.Model Model instance containing components and parameter structure. Modified in-place to reflect current parameter values. dim : int Dimensionality of spectrum to generate: - 1: Generate 1D spectrum (energy-resolved or time-resolved) - 2: Generate 2D spectrum (time- and energy-resolved) Returns ------- ndarray or list of ndarray Generated spectrum or spectra: - If dim=1 and plot_sum=True: 1D array (sum of components) - If dim=1 and plot_sum=False: List of 1D arrays (individual components) - If dim=2: 2D array (time x energy), regardless of plot_sum Examples -------- >>> # During fitting (1D) >>> spectrum = fit_model_mcp(energy, par_values, True, model, 1) >>> residual = data - spectrum >>> # For visualization (1D, individual components) >>> components = fit_model_mcp(energy, par_values, False, model, 1) >>> for i, comp in enumerate(components): ... plt.plot(energy, comp, label=f'Component {i}') >>> # During fitting (2D) >>> spectrum_2d = fit_model_mcp(energy, par_values, True, model, 2) >>> residual_2d = data_2d - spectrum_2d Notes ----- **Function Signature:** The signature follows the standard form [x, par, plot_sum, args] required by fitlib.residual_fun. The 'args' tuple contains (model, dim). **Parameter Update:** This function updates model.lmfit_pars in-place via model.update_value(). The model retains these values after the function returns. **2D Behavior:** For 2D models, plot_sum is ignored and the full 2D spectrum is always returned. Individual component plotting for 2D is typically done by examining time slices. **Performance:** 2D spectrum generation can be slow for large grids or complex models with many time-dependent parameters. Consider: - Reducing time/energy grid density during initial fits - Using fit_slice_by_slice for quasi-independent time points - Implementing parallel evaluation (model.create_value_2d_parallel) """ par_values: list[float] | np.ndarray if isinstance(par, np.ndarray): par_values = par else: par_values = list(par) model.update_value(new_par_values=par_values) # Update lmfit parameters # Create energy- (and time-)resolved spectrum/data if dim == 1: # 1D if not plot_sum: # Return individual components model.create_value_1d(store_1d=1) return model.component_spectra # Return sum of all components model.create_value_1d() if model.value_1d is None: raise RuntimeError("Model evaluation did not produce value_1d") return model.value_1d if dim == 2: # 2D model.create_value_2d() if model.value_2d is None: raise RuntimeError("Model evaluation did not produce value_2d") return model.value_2d raise ValueError(f"Unsupported dim={dim}; expected 1 or 2")
# @overload def fit_model_gir( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: Literal[True], *args: Any, ) -> np.ndarray: ... @overload def fit_model_gir( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, *args: Any, ) -> np.ndarray | list[np.ndarray]: ...
[docs] def fit_model_gir( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, *args: Any, ) -> np.ndarray | list[np.ndarray]: """Generate spectrum using the compiled GIR backend when available. When the first element of *args* is a compiled plan the fast evaluator is used. Otherwise the call is forwarded to :func:`fit_model_mcp`. Parameters ---------- x : array-like Independent variable axis (energy or time). par : array-like Full parameter vector (all params, fixed + varying). plot_sum : bool Component return mode. For 1D compiled plans, ``plot_sum=False`` falls back to the interpreter for component extraction. *args Either ``(plan, theta_indices, model, dim)`` for the compiled path, or ``(model, dim)`` for the interpreter fallback. The compiled path always carries *model* and *dim* so the interpreter is available when needed (e.g. component extraction). """ if isinstance(args[0], (ScheduledPlan2D, ScheduledPlan1D)): plan = args[0] theta_indices: np.ndarray = args[1] model: Model = args[2] dim: int = args[3] # 1D component extraction requires the interpreter. if isinstance(plan, ScheduledPlan1D) and not plot_sum: return fit_model_mcp(x, par, plot_sum, model, dim) par_arr = np.asarray(par, dtype=np.float64) theta = par_arr[theta_indices] if isinstance(plan, ScheduledPlan2D): return evaluate_2d(plan, theta) return evaluate_1d(plan, theta) # Non-lowerable fallback return fit_model_mcp(x, par, plot_sum, *args)
# @overload def fit_model_jax( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: Literal[True], *args: Any, ) -> np.ndarray: ... @overload def fit_model_jax( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, *args: Any, ) -> np.ndarray | list[np.ndarray]: ...
[docs] def fit_model_jax( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, *args: Any, ) -> np.ndarray | list[np.ndarray]: """Generate spectrum using a compiled JAX evaluator when available. When the first element of *args* is a callable it is the jitted evaluator from ``eval_jax.make_evaluator_2d_jax``; otherwise the call is forwarded to :func:`fit_model_gir` (NumPy compiled plan or interpreter fallback). Parameters ---------- x : array-like Independent variable axis (energy or time). par : array-like Full parameter vector (all params, fixed + varying). plot_sum : bool Component return mode (2D JAX path always returns the sum). *args ``(evaluator, jacobian, theta_indices, model, dim)`` for the JAX path — *jacobian* is carried for ``fitlib.jacobian_fun`` (lmfit ``Dfun``), not used here. Otherwise the :func:`fit_model_gir` conventions apply. Notes ----- The evaluator/jacobian entries are per-plan closures and do not pickle; MCMC via ``lmfit.emcee`` with ``workers > 1`` is not supported on this path (single-worker MCMC works). """ if callable(args[0]): evaluator = args[0] theta_indices: np.ndarray = args[2] par_arr = np.asarray(par, dtype=np.float64) return np.asarray(evaluator(par_arr[theta_indices])) return fit_model_gir(x, par, plot_sum, *args)
# @overload def fit_model_compare( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: Literal[True], *args: Any, ) -> np.ndarray: ... @overload def fit_model_compare( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, *args: Any, ) -> np.ndarray | list[np.ndarray]: ...
[docs] def fit_model_compare( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, *args: Any, ) -> np.ndarray | list[np.ndarray]: """Run both GIR and interpreter paths, compare results. When the first element of *args* is a compiled plan (:class:`ScheduledPlan2D` or :class:`ScheduledPlan1D`) the fast path is executed and its output compared against the interpreter via ``np.testing.assert_allclose``. On fallback the interpreter is called directly. Parameters ---------- x : array-like Independent variable axis (energy or time). par : array-like Full parameter vector (all params, fixed + varying). plot_sum : bool Component return mode. *args ``(plan, theta_indices, model, dim)`` for the comparison path, or ``(model, dim)`` for interpreter-only fallback. """ if isinstance(args[0], (ScheduledPlan2D, ScheduledPlan1D)): theta_indices: np.ndarray = args[1] model: Model = args[2] dim: int = args[3] fast = fit_model_gir(x, par, plot_sum, args[0], theta_indices, model, dim) slow = fit_model_mcp(x, par, plot_sum, model, dim) np.testing.assert_allclose(fast, slow, rtol=1e-10, atol=1e-10) return fast return fit_model_mcp(x, par, plot_sum, *args)
#
[docs] def fit_project_mcp( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, project_fit_info: dict[str, Any], dim: int, ) -> np.ndarray: """ Generate concatenated spectra from multiple files for project-level fitting. Distributes combined optimizer parameters to individual file models, evaluates each model, slices to each file's fit region, and returns one concatenated array for residual computation. Parameters ---------- x : array-like Unused (kept for fit-function signature compatibility). par : list or array-like Combined parameter values proposed by the optimizer, ordered to match ``project_fit_info["par_names"]``. plot_sum : bool Unused (kept for signature compatibility). Always returns sum. project_fit_info : dict Fitting context built by ``Project._build_fit_params()``: - ``"mapping"``: list of ``(project_name, file_idx, local_name)`` - ``"files"``: list of File objects (limits read from ``f.e_lim``/``f.t_lim``) - ``"models"``: list of Model objects (one per file) - ``"par_names"``: list of combined parameter names dim : int Must be 2 (project-level fitting is 2D only). Returns ------- ndarray Concatenated (flattened) fit arrays from all files. """ par_values = list(par) if not isinstance(par, np.ndarray) else par mapping = project_fit_info["mapping"] files = project_fit_info["files"] models = project_fit_info["models"] par_names = project_fit_info["par_names"] # Build name→value lookup from the combined parameter vector par_lookup: dict[str, float] = {} for i, name in enumerate(par_names): par_lookup[name] = float(par_values[i]) # Distribute values to each file's model for project_name, file_idx, local_name in mapping: model = models[file_idx] if local_name in model.lmfit_pars: model.lmfit_pars[local_name].value = par_lookup[project_name] # Evaluate each file and collect sliced results slices: list[np.ndarray] = [] for i, (model, f) in enumerate(zip(models, files, strict=True)): model.create_value_2d() if model.value_2d is None: raise RuntimeError( f"Model evaluation for file {i} did not produce value_2d" ) fit_2d = model.value_2d # Apply per-file slicing (read limits from file directly) if f.e_lim and f.t_lim: fit_2d = fit_2d[f.t_lim[0] : f.t_lim[1], f.e_lim[0] : f.e_lim[1]] elif f.e_lim: fit_2d = fit_2d[:, f.e_lim[0] : f.e_lim[1]] elif f.t_lim: fit_2d = fit_2d[f.t_lim[0] : f.t_lim[1], :] slices.append(fit_2d.flatten()) return np.concatenate(slices)
#
[docs] def pack_project_theta( plans: Sequence[ScheduledPlan2D], *, mapping: list[tuple[str, int, str]], par_names: list[str], var_names: list[str], ) -> tuple[np.ndarray, list[np.ndarray]]: """Pack the combined-parameter mapping into gather index arrays. Converts the name-based parameter distribution of project-level fitting (``project_fit_info["mapping"]`` from ``Project._build_fit_params``) into integer index arrays so the fused evaluator can scatter the combined optimizer vector to per-file plan thetas with plain array gathers — no name lookups per residual call. Parameters ---------- plans : sequence of ScheduledPlan2D One compiled plan per file, in project file order. Each plan's ``opt_param_names`` are local (per-file) parameter names. mapping : list of tuple ``(combined_name, file_idx, local_name)`` triples from ``Project._build_fit_params``. par_names : list of str All combined parameter names, in combined-vector order. var_names : list of str Varying combined parameter names, ordered as in ``par_names``. Defines the combined theta vector ``theta_c``. Returns ------- theta_c_indices : ndarray Positions of ``var_names`` within ``par_names``: ``theta_c = par_full[theta_c_indices]``. plan_gathers : list of ndarray Per file, positions within ``theta_c`` of that plan's opt params: ``theta_f = theta_c[plan_gathers[file_idx]]`` yields the plan's theta in ``opt_param_names`` order. Raises ------ RuntimeError If a plan opt param has no combined counterpart, its counterpart is not varying, or a varying combined param feeds no plan — all indicate plans built with vary flags that do not match the combined parameter set. """ remaps: list[dict[str, str]] = [{} for _ in plans] for combined_name, file_idx, local_name in mapping: remaps[file_idx][local_name] = combined_name full_pos = {name: i for i, name in enumerate(par_names)} var_pos = {name: i for i, name in enumerate(var_names)} theta_c_indices = np.array([full_pos[name] for name in var_names], dtype=np.intp) consumed: set[str] = set() plan_gathers: list[np.ndarray] = [] for file_idx, plan in enumerate(plans): remap = remaps[file_idx] gather: list[int] = [] for local_name in plan.opt_param_names: mapped_name = remap.get(local_name) if mapped_name is None: raise RuntimeError( f"Plan opt param '{local_name}' (file {file_idx}) has " f"no combined-parameter mapping entry." ) pos = var_pos.get(mapped_name) if pos is None: raise RuntimeError( f"Plan opt param '{local_name}' (file {file_idx}) maps " f"to combined param '{mapped_name}', which is not " f"varying." ) gather.append(pos) consumed.add(mapped_name) plan_gathers.append(np.array(gather, dtype=np.intp)) leftover = [name for name in var_names if name not in consumed] if leftover: raise RuntimeError( f"Varying combined params feed no plan: {leftover}. " f"Plans were likely built with stale vary flags." ) return theta_c_indices, plan_gathers
#
[docs] def fit_project_jax( x: Sequence[float] | np.ndarray, par: Sequence[float] | np.ndarray, plot_sum: bool, *args: Any, ) -> np.ndarray: """Generate concatenated multi-file prediction via the fused JAX evaluator. Project-fit counterpart of :func:`fit_model_jax`. The fused evaluator applies per-file fit windows, flattens, and concatenates internally, so the return value aligns element-for-element with the concatenated data vector assembled by ``Project.fit_2d`` — no slicing in ``fitlib.residual_fun`` (empty ``e_lim``/``t_lim``). Unlike ``fit_model_jax`` there is no fallback branch: ``Project`` dispatches here only when every file's graph passed the JAX gate; otherwise the const carries ``fit_project_mcp``. Parameters ---------- x : array-like Unused (kept for fit-function signature compatibility). par : array-like Full combined parameter vector (varying + static + expr), in ``project_fit_info["par_names"]`` order. plot_sum : bool Unused (the fused path always returns the sum). *args ``(evaluator, jacobian, theta_c_indices, var_names, dim)`` — *evaluator*/*jacobian* from ``eval_jax.make_project_evaluator_2d_jax`` / ``make_project_jacobian_2d_jax``; *jacobian* and *var_names* are carried for ``fitlib.jacobian_fun_project`` (lmfit ``Dfun``), not used here. Notes ----- The evaluator/jacobian entries are fused closures and do not pickle; MCMC via ``lmfit.emcee`` with ``workers > 1`` is not supported on this path (single-worker MCMC works). """ evaluator = args[0] theta_c_indices: np.ndarray = args[2] par_arr = np.asarray(par, dtype=np.float64) return np.asarray(evaluator(par_arr[theta_c_indices]))