Running example on home page (and dependencies)

[willingc]

Hi @tlambert03, I was playing around with ndv, pygfx, and uv.

I installed using the command from your docs' install command generator:

uv venv
source .venv/bin/activate
uv pip install "ndv[pygfx,pyqt]"

I tried to run the simple example on the home page. I needed to install a few more dependencies:

uv pip install "imageio[tifffile]"

Worked fine after that.

---

Interestingly, the command on the homepage using vispy pip install ndv[qt] does install the imageio library.

[tlambert03]

yeah that's not a terribly welcoming experience is it? 😂

Maybe I can tell you a few random details about ndv (and the various extras/examples), and you can help me decide how to present it all (or even if I should change some of my extras)

• first, we do have packaged data that requires no additional deps:

  uv venv
  source .venv/bin/activate
  uv pip install "ndv[pygfx,pyqt]"
  python -c "import ndv; ndv.imshow(ndv.data.nd_sine_wave())"

  so, we could just use ndv.data.nd_sine_wave() for that initial documentation (but it's kinda boring 😆). In any case, i absolutely agree that it's silly to have the very first example not work without a special install

• in an attempt to be generally compatible but with bare minimal dependencies... we have (for better or worse) leaned heavily on extras, as you've seen. The jupyter, pyqt, pyside and wxpython extras, bring in just a gui framework, but no canvas. The vispy and pygfx extras bring in just a graphics framework... But then I also made these somewhat opinionated "ready to go bundled" extras": qt, jup, wx... which, as you've found, also bring both a canvas backend (vispy by default) and the io dependencies. (i guess those are analogous to napari[all]).

• all the examples in the repo (https://github.com/pyapp-kit/ndv/tree/main/examples) have PEP-723-style inline deps... so you could run, for example:

  uv run https://github.com/pyapp-kit/ndv/raw/refs/heads/main/examples/numpy_arr.py

  or any other example like that. (but I haven't really mentioned that in the docs) ... and in the numpy_arr.py example, you'll see I even try/catch the ndv.data.cells3d() import 😂

---

Anyway, sorry to make your simple point a more complex one, but do you have any general opinions on having the initial example be an interesting/compelling one, using real-looking data, without adding imageio[tifffile] to my base dependencies? or just use the sine way example?

[willingc]

I agree that the sine wave is too boring. 😄

Let me noodle a bit on the rest.

[tlambert03]

One possibility is of course to just roll my own. Store a few npz files online somewhere and download them without imageio or tifffile. That would be easy enough and provide nice data without more deps