Source code for suboptimumg.log_analysis.utils

import numpy as np
import numpy.typing as npt




def safe_gradient(
    signal: npt.NDArray[np.float64], dt: npt.NDArray[np.float64]
) -> npt.NDArray[np.float64]:
    """Compute a time derivative that is robust to zero-width time steps,
    so that duplicate timestamps do not produce inf or NaN values.

    Parameters
    ----------
    signal : npt.NDArray[np.float64]
        The signal to differentiate.
    dt : npt.NDArray[np.float64]
        Time step array. Must be the same length as ``signal``.

    Returns
    -------
    npt.NDArray[np.float64]
        ``d(signal)/dt`` with NaNs from zero-width steps filled by
        linear interpolation between the nearest valid neighbours.
    """
    # Compute the raw gradient, masking out any points where dt == 0
    # to avoid division-by-zero producing inf or incorrect values.
    result = np.full_like(signal, np.nan)
    mask = dt != 0
    result[mask] = np.gradient(signal[mask]) / dt[mask]

    nans = np.isnan(result)
    if nans.any():
        # Fill masked positions by linearly interpolating from valid neighbours.
        idx = np.arange(len(result))
        result[nans] = np.interp(idx[nans], idx[~nans], result[~nans])

    return result