kinextract.fit_state_map_with_optional_clean

kinextract.fit_state_map_with_optional_clean(st: FitState, cfg: FitConfig, a0: ndarray, bounds: list)[source]

Run the MAP fit, optionally with sigma-clipping.

This is the shipped (non-joint) fit path: a bound-constrained L-BFGS-B optimisation of chi2 + wing-tapered smoothness penalty + LOSVD normalisation penalty (kinextract.numerics.objective_map()), the same objective minimised by the original Fortran implementation this package is a port of. Used directly by pre-normalized-mode fits (cfg.fit_continuum=False) and internally by kinextract.errors’s residual-bootstrap error estimation, which needs hundreds to thousands of fast independent refits per fit – a cost only a point-estimate optimisation, not full posterior sampling, can sustain. A full-posterior (NUTS/HMC) alternative is available via kinextract.bayesian.fit_state_bayesian() for users who want a sampled posterior instead of a point estimate; it is not used by default because a comprehensive recovery-accuracy comparison found the MAP point estimate matches or exceeds the posterior mean’s LOSVD-shape recovery across tested instruments and LOSVD shapes, at a small fraction of the runtime.

This is the mid-level orchestration function: it optionally runs the automatic xlam grid search (_auto_select_xlam()) once up front, then performs a single (optionally sigma-clipped) MAP fit. Continuum-cofit fits (cfg.fit_continuum=True) never reach this function – run_spectral_fit() dispatches those to kinextract.joint instead.

Parameters:

  • st (FitState) – Fit state to optimise; st.xlam may be overwritten by the automatic regularisation search, and st.clean_good_mask is updated as a side effect when cleaning is enabled.

  • cfg (FitConfig) – Run configuration controlling regularisation search (xlam_auto and related settings) and cleaning (clean and related settings).

  • a0 (ndarray) – Initial guess for the flat parameter vector.

  • bounds (list of tuple) – Per-parameter (lower, upper) bounds.

Returns:

The optimisation result from the MAP fit (after any cleaning).

Return type:

scipy.optimize.OptimizeResult