08 1-D Arrays

imcombiners.kernels can also work on ordinary 1-D value vectors. The _1d functions expose the same NaN-aware reductions and rejection semantics used by the image-stack kernels, without requiring you to reshape a vector into (N, 1, 1) yourself.

import numpy as np
import imcombiners.kernels as imck

Scalar Statistics

Use the direct scalar functions for generic vectors, flattened images, light curves, or detector samples:

rng = np.random.default_rng(20260528)
values = rng.normal(1000.0, 30.0, size=10_000).astype("float32")
values[0] = np.nan
values[1] += 700.0

print(imck.mean_1d(values))
print(imck.median_1d(values))
var, mean = imck.var_1d(values, return_mean=True)
print(var, mean)

1000.5888671875
1000.76025390625
948.0440063476562 1000.5888671875

Names use Python’s common words for 1-D utilities:

Function	Meaning
`mean_1d(values)`	NaN-aware mean
`median_1d(values)`	NaN-aware median
`lmedian_1d(values)`	lower median, matching IRAF/imcombine convention
`sum_1d(values)`	NaN-aware sum
`min_1d(values)`	NaN-aware minimum
`max_1d(values)`	NaN-aware maximum
`var_1d(values, ddof=0, return_mean=False)`	NaN-aware variance; optionally returns `(variance, mean)`
`wvg_1d(values, weights)`	NaN-aware weighted average

Rejection Masks

The rejection utilities return the same products as stack rejection, but with 1-D shapes and scalar diagnostics:

mask_rej, std, low, upp, nit, flags = imck.sigclip_1d(
    values,
    sigma=3.0,
    maxiters=5,
    cenfunc="median",
)

print(mask_rej.shape)
print(std, low, upp, nit, flags)

(10000,)
29.535048 912.39746 1088.95 4 0

The mask-only functions are convenient when you want to inspect or reuse the survivors yourself:

mask = imck.sigclip_mask_1d(values, sigma=3.0)
survivors = values[~mask & np.isfinite(values)]

print(survivors.size)
print(imck.median_1d(survivors))

9970
1000.7613525390625

Rejected Combine

For output-only robust statistics, call the _combine_1d helpers:

sigclip_median = imck.sigclip_combine_1d(
    values,
    combine="median",
    sigma=3.0,
    maxiters=5,
)

trimmed_mean = imck.minmax_combine_1d(
    values,
    combine="mean",
    n_min=1,
    n_max=1,
)

print(sigclip_median, trimmed_mean)

1000.7613525390625 1000.5289916992188

The available rejection families mirror the image-stack API:

Full result	Mask only	Output-only combine
`sigclip_1d`	`sigclip_mask_1d`	`sigclip_combine_1d`
`ccdclip_1d`	`ccdclip_mask_1d`	`ccdclip_combine_1d`
`minmax_1d`	`minmax_mask_1d`	`minmax_combine_1d`
`pclip_1d`	`pclip_mask_1d`	`pclip_combine_1d`
`linearclip_1d`	-	-

Validation and Hot Loops

By default the functions validate shape, dtype, mask length, and contiguity. In a tight loop, prepare the vector once and pass validate=False:

workspace = np.ascontiguousarray(values, dtype=np.float32)
out = imck.median_1d(workspace, validate=False)
print(out)

1000.76025390625

Use validate=False only when you already know the input is a 1-D contiguous float32 or float64 array and any mask/weights have matching length. The validated path is the safer default for scripts and notebooks.