coshape

Shape-preserving 1-D spline interpolation, implemented in Rust.

Given concave data the fitted spline passes through the knots exactly, keeps f'' <= 0 everywhere, and is at least C². Convex data gives the mirror image; the tension and variable-degree methods also follow data that changes curvature.

import array
from coshape import EndSlopes, Tension

x = array.array("d", [0.0, 1.0, 2.0, 3.0, 4.0])    # any float64 buffer, e.g. numpy
y = array.array("d", [0.0, 4.0, 7.0, 7.5, 7.9])    # concave

fitter = Tension().with_boundary(EndSlopes.clamped(4.5, 0.3))   # configuration only
spline = fitter.fit(x, y)                                     # a separate, fitted object

spline(2.5)                           # value
spline.second_derivative(2.5)         # <= 0 on concave data

grid = array.array("d", (i / 100 for i in range(401)))
spline.value_batch(grid)                   # a new array.array('d')
out = array.array("d", bytes(8 * len(grid)))
spline.value_batch_into(grid, out)         # fills your buffer, no allocation

Fitters are immutable: each with_* returns a new fitter. Arrays go through the buffer protocol — numpy float64 arrays, array.array('d'), or memoryviews of them — with no numpy dependency, which is why Python 3.11 or newer is required. Invalid data raises ValueError. Type stubs are included.

Dual-licensed under MIT OR Apache-2.0. Source: https://github.com/Feiyang472/coshape