savgoldiff uses axis 0 length when differentiating another axis

[alexzerg]

Now it's time for multidimensional input with time on axis 1. Here x has shape (2, 41) and time is on axis 1:

import numpy as np
from pynumdiff.polynomial_fit import savgoldiff

t = np.linspace(0, 2*np.pi, 41)
dt = t[1] - t[0]

x = np.vstack([
 np.sin(t),
 np.cos(t),
])

x_hat, dxdt_hat = savgoldiff(
 x,
 dt,
 degree=3,
 window_size=11,
 smoothing_win=5,
 axis=1,
)

Oops (joss2.0 python 3.13)...
ValueError: polyorder must be less than window_length

Same data transposed to (41, 2) works with axis=0. Does it take window limits from len(x) so here it sees 2 samples instead of 41? Probably should use the selected axis length.

Review reference: openjournals/joss-reviews#11172

[pavelkomarov]

I was using len(x) instead of x.shape[axis], a holdover from the pre-multidimensional era.

[pavelkomarov]

24cd55f