Surface plots

[anetasie]

I was just looking at the images that were made with a very old version of Sherpa for the Chandra Newsletter in 2001. I'm wondering if we are able to easily make such plots now and also with a better quality than in the past.

[DougBurke]

Technically we can, as long as the backend allows it

• matplotlib does - e.g. https://matplotlib.org/stable/plot_types/3D/surface3d_simple.html
• bokeh I don't think does directly, but there are additions that could, but I could be very wrong

This is related to

• 1517

The issue becomes what the user-interface is (a new set of calls or allow plot_data to do it or ...) and then how do we "control" it (e.g. to rotate it and change the viewing angle and how the surface is rendered and ...).

[DougBurke]

We can explain to uses how to take the output of get_data_image/get_model_image etc and use it with matplotlib directly.

[DougBurke]

As an example, but be careful as I always get the ordering wrong (which is why I use a rectangular grid with an elliptical offset model)

dataspace2d([250, 200])
set_source(gauss2d.mdl)
mdl.xpos = 120
mdl.ypos = 70
mdl.fwhm = 50
mdl.ellip = 0.6
mdl.theta = 30 * np.pi / 180

image_model()

# access to coordinate information is not as simple as the 1D plot cases (e.g. get_data_plot) so we
# do it old-school, and hope we don't get things confused
m = get_model_image()
x = np.arange(1, 251)
y = np.arange(1, 201)
xg, yg = np.meshgrid(x, y)

from matplotlib import cm
fig, ax = plt.subplots(subplot_kw={"projection": "3d"})
ax.plot_surface(xg, yg, m.y, cmap=cm.viridis)
ax.set(xlabel="x", ylabel="y")

[hamogu]

We currently don't have a matplotlib image backend (images are shown in ds9). Once we have that, it's a trivial change to use surfaces instead. However, do we even want that? As @DougBurke said get_data_image plus a plotting backend of your choice can do that already. I don't think it's a good use of our time to make an interface for any and all plots that we can think of, but instead, we should provide a generic interface to make one usuable plot and leave it to the user to use functions like get_data_imge or to to inherit from out backends for other vizualization choices.

Today, I think the default way to display an image is color (as done in ds9 and as I would do it in a matplotlib-based backend for image plotting).

[anetasie]

@DougBurke thanks for the example. It looks great and much better quality than the old plot. I agree that we should just document this option for the users.

[anetasie]

btw. I will try with some images of residuals.