FitProcessor

The FitProcessor class encapsulates all curve fitting logic, independently of the UI. It handles multiple fitting strategies (lmfit, ODR, emcee) and supports 1D fitting, multi-1D (Fit per Y), and full 2D surface fitting.

Responsibilities

  • Convert a formula string to a Python function (generate_fit_function)
  • Prepare fitting parameters and options (prepare_parameters)
  • Dispatch the fitting logic depending on the selected method and strategy
  • Return results in a unified format: (result, minimizer, best_fit, confidence_band)

Entry Points

1. Standard 1D Fit

result, model, best_fit, band = processor.process_fit(
    formula="y = A * exp(-B * x)",
    x_data=x,
    y_data=y,
    weights=w,
    weighting_method="None",
)

Dispatches to:

  • lmfit → _process_lmfit()
  • odr → _process_odr()
  • emcee → _run_emcee()

2. Multi 1D Fit (Z = f(X) with fixed Y)

result, minimizer, best_fit, band = processor.process_fit_per_y(formula, x, y, z)

Performs a global fit with Y treated as a fixed value in each Z slice. Supports lmfit, odr, emcee.

3. Full 2D Surface Fit

func = processor.generate_fit_function(formula)
result, _, best_fit, band = processor.process_fit_surface(formula, func, x, y, z)

Z is modeled as a function of both X and Y. Each method is supported:

  • lmfit: process_fit_surface
  • odr: process_fit_surface_odr
  • emcee: process_fit_surface_emcee

MCMC Support (emcee)

All MCMC fits are handled via:

_run_emcee(func, param_info, inputs, targets, sigma, fit_options, mode)

The mode argument changes the likelihood evaluation:

  • "1d"
  • "per_y" (multi 1D)
  • "surface"

Returns:

result: dict
sampler: emcee.EnsembleSampler
best_fit: np.ndarray
confidence_band: Tuple[np.ndarray, np.ndarray]

Confidence Bands

The 99% confidence bands are:

  • computed from result.eval_uncertainty(sigma=3) in lmfit
  • extracted from the MCMC prediction percentiles
  • approximated via the covariance matrix in ODR

Method Summary

Method Role
generate_fit_function() Turns formula into executable Python function
prepare_parameters() Extracts p0, bounds from FitOptions dialog
process_fit() Entry for standard 1D fit
process_fit_per_y() Entry for multi-1D fit (Z = f(X), fixed Y)
process_fit_surface() Entry for full surface fit
_run_emcee() Shared backend for MCMC fits
_process_lmfit() Core least-squares fitting with lmfit.Model
_process_odr() Fitting using orthogonal distance regression (ODR)
reset_fit() Clears result, toolbar, and reinitializes plots