Skip to main content

kernels.minmax_combine_1d

kernels.minmax_combine_1d(
    values,
    *,
    mask=None,
    combine,
    n_min=1,
    n_max=1,
    validate=True,
)

Reject the n_min smallest and n_max largest unmasked values at each pixel. Single-pass (1-D output-only variant).

Parameters

values : (ndarray, shape(N))

One-dimensional value vector. Accepted dtypes are uint8, uint16, int16, int32, float32, and float64. Integer inputs are promoted to the package’s floating workspace when validate is True.

mask : ndarray of bool = None

Input mask; True means already masked. Must have shape (N,).

combine : {'mean', 'median'}

Output combine method evaluated after rejection. Fused combine kernels support "mean", "average", "avg", "median", and "med".

n_min : int or float = 1

Number of minimum-side values to reject at each pixel. Values >= 1 are treated as a frame count. Values in [0, 1) are treated as a fraction of the total frame count N and converted via int(N * n_min + 0.001), matching IRAF’s internal fraction arithmetic.

n_max : int or float = 1

Same convention as n_min, applied to the maximum-side tail.

validate : bool = True

If True, check dimensionality and normalize dtype/contiguity before entering the Rust kernel. If False, callers must provide inputs that satisfy the compiled kernel assumptions.

Returns

combined : scalar

Mean or median of the surviving values after applying the input mask, finite-value filtering, and this rejection algorithm.

Notes

Sort the M finite, unmasked samples at each output pixel. With N the original image count, a request a (n_min or n_max) resolves to::

A = int(N * a + 0.001) if 0 <= a < 1 else int(a)
k = int(M * A / N + 0.001)

Reject k samples from that tail, first the low tail, then the high tail of the remaining values. Both counts use the initial M; available samples limit the cuts. Thus masked/non-finite inputs reduce even count-based requests proportionally. For example, N=10, M=5, n_min=2 removes one minimum. This is rank selection, not a value threshold: tied values may be split. There is no spread estimate or iteration; low/upp are retained extrema. For derivations and worked examples, see the Rejection Methods tutorial.