How to use in Google Colab?

[ziw-liu]

How can I use this in Google Colab notebooks? I tried to follow the Jupyter example:

!pip install "ndv[jup]"

import ndv

data = ndv.data.cells3d()
ndv.imshow(data)

and got an error:

---------------------------------------------------------------------------

RuntimeError                              Traceback (most recent call last)

[<ipython-input-2-2c8bdd59b309>](https://localhost:8080/#) in <cell line: 0>()
      2 
      3 data = ndv.data.cells3d()
----> 4 ndv.imshow(data)

3 frames

[/usr/local/lib/python3.11/dist-packages/ndv/views/_app.py](https://localhost:8080/#) in ndv_app()
    210             errors.append((key, e))
    211 
--> 212     raise RuntimeError(  # pragma: no cover
    213         f"Could not find an appropriate GUI frontend: {valid!r}. Tried:\n\n"
    214         + "\n".join(f"- {key}: {err}" for key, err in errors)

RuntimeError: Could not find an appropriate GUI frontend: {'jupyter', 'wx', 'qt'}. Tried:

- GuiFrontend.QT: No module named 'qtpy'
- GuiFrontend.WX: No module named 'wx'
- GuiFrontend.JUPYTER: Jupyter is not running a notebook shell.  Cannot create app.

[ziw-liu]

Copying the check from:

ndv/src/ndv/views/bases/_app.py

Lines 63 to 67 in 371a878

	@staticmethod
	def _ipython_shell() -> InteractiveShell | None:
	if (ipy := sys.modules.get("IPython")) and (shell := ipy.get_ipython()):
	return cast("InteractiveShell", shell)
	return None

import sys
from typing import cast
from IPython.core.interactiveshell import InteractiveShell

if (ipy := sys.modules.get("IPython")) and (shell := ipy.get_ipython()):
    print(cast("InteractiveShell", shell))

<google.colab._shell.Shell object at 0x782776d01f10>

And it fails this condition:

ndv/src/ndv/views/_jupyter/_app.py

Line 27 in 371a878

return bool(ipy_shell.__class__.__name__ == "ZMQInteractiveShell")

[ziw-liu]

After tinkering a little it seems like the Colab Shell is setup differently and GL can be tricky. Feel free to close if this is out of scope.

[tlambert03]

Yeah, I played with colab a month ago to see if I could get it to work. It's not impossible, nor is it out of scope. But it's still not turnkey I'm afraid. The best hope would likely be with pygfx (rather than vispy), since that doesn't require opengl... and is a bit easier to use with an offscreen buffer. So, you'd use pip install "ndv[pygfx,jupyter]" ... however that also isn't yet working, and if I recall correctly, it had more to do with idiosyncrasies of using jupyter widgets in a colab environment. I definitely want to make this work, so thanks for opening. It will likely require a little digging into the specifics of the colab environment though.

[gselzer]

I definitely want to make this work, so thanks for opening.

Ditto!