redim() of xarray causes contours to be off

[laleph]

Redim() should be the correct approach to alter the dimensions, right?

First I generate some test data.

#### TEST
import numpy as np
import xarray as xr

def gauss2d(amp, xo, yo, sigma_x, sigma_y, xy, xdim, ydim):
    """2d Gaussian """
    
    x, y = xy

    model = amp * np.exp( -(x-xo)**2 / (2*sigma_x**2) - (y-yo)**2 / (2*sigma_y**2)) 
#     return xr.DataArray(model, coords=[xdim, ydim], dims=['xdim', 'ydim'])
    return xr.DataArray(model, coords=[ydim, xdim], dims=['ydim', 'xdim'])

amp = 50
xo = -0.08
yo = -0.04  # AXES SWAPPED?!?
sigma_x = 0.03
sigma_y = 0.03

x = np.linspace(-0.15,0.15,91)
y = np.linspace(-0.15,0.15,91)
x2, y2 = np.meshgrid(x,y)

model = gauss2d(amp, xo, yo, sigma_x, sigma_y, (x2, y2), x, y)

Then I plot without redim() and the output is fine.

# PLOTS CORRECTLY
import holoviews as hv
hv.notebook_extension('matplotlib',width=90)

%opts Image (cmap="viridis") [colorbar=True, fontsize={'ticks':18, 'title':22, 'ylabel':20, 'xlabel':20, 'legend':18, 'legend_title':20}] {+axiswise}
%output size=180 dpi=90

img = hv.Image(model)
img * hv.operation.contours(img)

With redim() the contours are off.

# CONTOURS OFF, when redim() is used
lim = 0.1
img = hv.Image(model).redim.range(xdim=(-lim,lim), ydim=(-lim,lim))
img * hv.operation.contours(img)

The center of the contours is wrong in both dimensions.
This is probably related to #2319.

And the axes seem to be switched. Or am I mistaken? The shifted center of the gaussian has swapped the x and y coordinates. Or at least the labelling is confusing when xarray is used.

[laleph]

Apparently, the dimensions in xarray have a different ordering.

import matplotlib.pyplot as plt

model.plot.contourf()
plt.show()

I think that is confusing.

Nonetheless, the contour operation in holoviews has another issue.

[philippjfr]

Thanks @laleph, definitely a bug. Should hopefully be an easy fix.