hispecdrp.primitives.utils.ccf.ccf_gauss_fit_numba

Numba-parallel Gaussian (+ fixed baseline) CCF centroid for line-mask chunk RVs.

Replaces the serial Astropy fit_ccf_rv() loop in rv_per_chunks_line_mask() with a prange over chunks. Model matches the line-mask stored (peak-up) convention: fixed baseline + positive-amplitude Gaussian; two velocity passes; optional inverse-variance weights from ccf_err.

Functions

_fit_ccf_rv_rows_numba_impl(v_grid, ...)

_fit_one_ccf_row(v_grid, ccf, ccf_err, ...)

Peak-up CCF row → (rv_m_s, rv_err_m_s, ok).

_gauss(v, amp, mu, sig, baseline)

_lm_gauss_fixed_baseline(v, y, w, use_w, ...)

Weighted LM for y = A·exp(-½((v-μ)/σ)²) + B with fixed B.

_mad_1d(a, med)

_median_positive_1d(w)

Median of finite w > 0; 1 if none (weight renormalization).

_nanmedian_1d(a)

_sigma_clipped_median(y, n_sigma, max_iter)

_window_gauss_fit(v_grid, ccf, ccf_err, ...)

Fit Gaussian in |v-center|<=hw.

fit_ccf_rv_rows_numba(v_grid_m_s, ccf_rows, *)

Parallel Gaussian centroids for peak-up CCF rows.

fit_ccf_rv_rows_numba(v_grid_m_s, ccf_rows, *, ccf_err_rows=None, gaussian_seed_mean_m_s=None, pass1_half_width_m_s=800000.0, pass2_half_width_m_s=120000.0, initial_stddev_m_s=5000.0, levmar_maxiter=450, min_points=5, max_cores=None, pix_size_m_s=None)[source]

Parallel Gaussian centroids for peak-up CCF rows.

Parameters:
  • v_grid_m_s – Shared velocity grid (m s⁻¹), shape (n_v,).

  • ccf_rows – CCF values, shape (n_chunk, n_v), peak-up (absorption as positive lobe).

  • ccf_err_rows – Optional per-sample σ, same shape as ccf_rows.

  • gaussian_seed_mean_m_s – Optional discrete-peak seed center (m s⁻¹).

  • pix_size_m_s – Detector pixel width in velocity (m s⁻¹). When set, σ_μ is inflated by √nsp with nsp = max(pix_size / Δv_grid, 1) so oversampled CCF bins that share pixels are not treated as independent.

  • max_cores – Numba thread cap for prange.

Returns:

rv, rv_err, ok – Arrays length n_chunk (ok is int64 0/1).