Interactive PyVista inside a Jupyter Notebook about pyvista-support

[GuillaumeFavelier]

I thought I had the panel code to only run if the code is being executed from a Jupyter notebook.

And yes, it works as expected. My bad, it was not clear. What I mean is, I had to use a workaround in my specific use case because I expect that any call to .show() will show the renderer (c.f. plot_sphere_1() scenario) even if it's not the cell's output.

[banesullivan]

Do you think there would be a way to integrate such a feature in PyVista

Probably - we should add something like this in pyvista/pyvista#300. My only hesitation is that the url and port have to be known, and they might not always be notebook_url="localhost:8888", port=0 so maybe we should ping the panel dev team.

Frankly, I think all notebook related dev should be paused until a few things move forward with itk-jupyter-widgets. Reference InsightSoftwareConsortium/itkwidgets#154 and waiting on:

• InsightSoftwareConsortium/itkwidgets#163
• InsightSoftwareConsortium/itkwidgets#166

Once those two issues are addressed, we will favor using itk-jupyter-widgets for notebook rendering directly instead of all the hacky solutions where a VTK render window is exported/serialized for VTKjs, X3D, or K3D.

[GuillaumeFavelier]

After more investigation I found that this version of plot_scene_1() is way simpler and doesn't require notebook_show() or any unnecessary tweaks anymore:

def plot_scene_1():
    from IPython.display import display
    S = pv.Sphere()
    plotter = pv.Plotter()
    plotter.add_mesh(S, color='white')
    disp = plotter.show(use_panel=True, auto_close=False)
    display(disp)
plot_scene_1()

[GuillaumeFavelier]

Actually if I modify plot_scene_1() like the following, it is displayed automatically:

def plot_scene_1():
    S = pv.Sphere()
    plotter = pv.Plotter()
    plotter.add_mesh(S)
    return plotter.show()
plot_scene_1()

So I work on integrating this idea in mne-python. Please don't hesitate to tell me if you know any way to do it better.

[banesullivan]

@GuillaumeFavelier: the plotter.show() call is actually returning a panel.pane.vtk.VTK object that can be displayed in the Jupyter cell output. Calling plotter.show() and not having its result be the cell's output will result in the panel-based interactive rendering not be shown.

I expect this behavior... however this might be considered a bug as a user might expect that any call to .show() will show the renderer. Also, this is inconsistent with a the static screenshot display in Jupyter notebooks (which returns None and displays the screenshot without it being the cells output):

from pyvista import examples
import pyvista as pv

mesh = examples.download_st_helens().warp_by_scalar()

plotter = pv.Plotter()
plotter.add_mesh(mesh)
output = plotter.show(use_panel=False)

Perhaps we need to collaborate with the panel developers to see if there is a way to display a panel object without it being a cell's output?

[GuillaumeFavelier]

Many thanks for the detailed explanations @banesullivan, now I understand better how it works and I think I can manage with this solution for now. Although I had to rely on the part of code detecting if the script is executed inside a notebook.

Perhaps we need to collaborate with the panel developers to see if there is a way to display a panel object without it being a cell's output?

That would wonderful of course!

[banesullivan]

Although I had to rely on the part of code detecting if the script is executed inside a notebook.

That's strange, what exactly do you mean? I thought I had the panel code to only run if the code is being executed from a Jupyter notebook.

[GuillaumeFavelier]

Actually after some more research I found a way to force panel to show the pane thanks to the show_server() function:

So I used the following function:

def notebook_show(disp, notebook_url="localhost:8888", port=0):
    from panel.viewable import show_server
    show_server(disp, notebook_url, port)

Do you think there would be a way to integrate such a feature in PyVista @banesullivan ? Eventually, I can open a discussion with the Panel developers to find a better/cleaner solution just in case.