Source code for hispecdrp.primitives.nonlinear_correction.nonlincorr_numba

from numba import njit, prange
from ..utils.array_utils import polyval1d
import numpy as np
from ...datamodels.dqflags import DQ_FLAGS
from ...utils.parallelization_utils import numba_thread_scope

__all__ = ['correct_nonlinearity']

NON_LINEAR_FLAG_uint8 = np.uint8(DQ_FLAGS["NON_LINEAR"])
DO_NOT_USE_FLAG_uint8 = np.uint8(DQ_FLAGS["DO_NOT_USE"])

@njit(nogil=True, parallel=True, cache=True)
def _correct_nonlinearity_numba(
    ramps : np.ndarray,
    dq_raw : np.ndarray,
    dq : np.ndarray,
    coeffs : np.ndarray,
    nonlin_max : np.ndarray,
    nonlin_dq : np.ndarray
):
    n_reads, ny, nx = ramps.shape
    for i in prange(ny):
        for j in range(nx):
            first_read = ramps[0, i, j]
            nl_corr_ramp = first_read + polyval1d(coeffs[:, i, j], ramps[:, i, j] - first_read)
            ramps[:, i, j] = nl_corr_ramp
            for k in range(n_reads):
                if ramps[k, i, j] > nonlin_max[i, j]:
                    dq_raw[k, i, j] |= (NON_LINEAR_FLAG_uint8 | DO_NOT_USE_FLAG_uint8)
            dq[i, j] |= nonlin_dq[i, j]


[docs] def correct_nonlinearity( ramps : np.ndarray, dq_raw : np.ndarray, dq : np.ndarray, coeffs : np.ndarray, nonlin_max : np.ndarray, nonlin_dq : np.ndarray, max_cores : int = 1, ): """ Apply non-linearity correction to up-the-ramps. Correction is done in place. The first read is subtracted before applying the polynomial correction, but it is added back after such that the output ramp still includes the bias. The dq arrays are updated in place to flag any reads that exceed the non-linearity limit. Parameters ---------- ramps 3D array of shape (Nreads, Ny, Nx) containing the ramps. Must be floating point. dq_raw 3D array of shape (Nreads, Ny, Nx) containing the raw DQ flags. Any reads beyong `nonlin_max` will be flagged as "non linear" and not trustworthy. dq 2D array of shape (Ny, Nx) containing the DQ flags to be updated with non-linearity flags. coeffs 3D array of shape (Ncoeffs, Ny, Nx) containing the polynomial coefficients of P(). such that, "corrected ramp (DN)" = P("measured ramp (DN)"). nonlin_max The maximum DN value for which the non-linearity correction is good to within 1%. nonlin_dq 2D array of shape (Ny, Nx) containing the DQ flags from the non-linearity model to be applied to all pixels. max_cores Maximum number of CPU cores to use. Default is 1. Returns ------- ramps The non-linearity corrected ramps (same as input `ramps` but modified in place). dq_raw The updated raw DQ array (same as input `dq_raw` but modified in place). dq The updated DQ array (same as input `dq` but modified in place). """ assert not np.issubdtype(ramps.dtype, np.integer), f"Ramps must be floating point, got {ramps.dtype}" with numba_thread_scope(max_cores): _correct_nonlinearity_numba(ramps, dq_raw, dq, coeffs, nonlin_max, nonlin_dq) return ramps, dq_raw, dq