Interpolation 2D with SmoothBivariateSpline

I get some strange results when using SmoothBivariateSpline.

I may use it in a wrong way (there is no example in a doc), so I’ve created simple example to reproduce my problem. Using known recipe for a 2d surface, I sample it and try to interpolate. To make it simple I use regularly spaced grid of points although the Spline is said to allow to use non-regular grid.

import matplotlib.pyplot as plt
import numpy as np
import scipy.interpolate

train_x, train_y = np.meshgrid(np.arange(-5, 5, 0.5), np.arange(-5, 5, 0.5))
train_x = train_x.flatten()
train_y = train_y.flatten()
z_func = lambda x, y: np.cos(x) + np.sin(y) ** 2

train_z = z_func(train_x, train_y)
interp_func = scipy.interpolate.SmoothBivariateSpline(train_x, train_y, train_z)

test_x = np.arange(-10, 10, 0.01)
test_y = np.arange(-10, 10, 0.01)
grid_x, grid_y = np.meshgrid(test_x, test_y)

interp_result = interp_func(test_x, test_y)
perfect_result = z_func(grid_x, grid_y)

fig, axes = plt.subplots(1, 2, figsize=(12, 8))
extent = [test_x[0], test_x[-1], test_y[0], test_y[-1]]
axes[0].imshow(perfect_result, aspect='auto', cmap='nipy_spectral', extent=extent)
axes[0].plot(train_x, train_y, 'w.')
axes[0].set_title('Perfect result, sampled function')
axes[1].imshow(interp_result, aspect='auto', cmap='nipy_spectral', extent=extent)
axes[1].plot(train_x, train_y, 'w.')
axes[0].set_title('SmoothBivariateSpline')
plt.show()

The result is somewhat strange:

Right image should be similar to left image in the middle area, where I’ve provided sample points of the surface.

Do I have some error? Or is there a bug? Or maybe you know of some other ways of interpolating using non-regularly spaced points?

[ev-br]

This example is now a part of the tutorial, merged as a part of gh-16707. Thanks @dankal444 !

[ev-br]

This example is now included into gh-16707. Thank you @dankal444 ! A review of gh-16707 from a user perspective would be much appreciated BTW.