fix: make sure pillow is not partially upgraded - #7

@psschwei I don't think this was solvable via UV - hence the sys.executable call. However in doing some additional investigating around this I found the best way to fix this seems to be just restarting the kernel to pickup updates instead of dealing with a stale cached version (being served via the colab env instead of uv).

I tried to run the ICML notebook in collab and it crashed on me running the first cell a couple of times. Here's the logs:

I'm not really sure what to make of these errors tbh.

Also will tag in @jakelorocco and @nrfulton in case they have an alternative viewpoint on whether we should try to maintain these notebooks on collab...

Did you try running sequential cells after the "crash"? This IPython.Application.instance().kernel.do_shutdown(True) - which makes sure we don't have any cache fun, also pops up the little crash helper even though the cell ran fine. I assume it's detecting the kernel being down as a crash. Running the following cells works as expected.

May be a better way of doing it though.

Edit: assuming you also have the zstd fix in #6

I did not try running additional cells.

Is there any way to avoid or hide the warnings? Feels like that would be a better UX if possible

I did some digging and could not find a way to restart it without the warning or suppress it.

Also will tag in @jakelorocco and @nrfulton in case they have an alternative viewpoint on whether we should try to maintain these notebooks on collab...

Personally, I think the notebooks on collab are useful when doing demos / presentations. If the errors / difficulties are different between the notebooks, we should at least have a current that we try to maintain / minimize errors for.

we should at least have a current that we try to maintain / minimize errors for.

I like this idea

Errors seem to be the same. I think the question is, does less than graceful kernel restart in cell 1 (to make sure the new version of updated dependencies are picked up instead of accidentally using old colab base ones) a deal breaker? Or as long as we add a comment about Colab reporting a kernel crash would that be fine?

I haven't found another avenue to address the issue without causing the crash popup or a way to suppress it.

If not a deal breaker should we put it into all of the notebooks or just make a new one for current and leave the old ones as is? cc @jakelorocco

If not a deal breaker should we put it into all of the notebooks or just make a new one for current and leave the old ones as is?

My two cents: I would make a new one. For the old ones, I'd be tempted to retire them and have them redirect to the new one (mainly so we can catch anyone coming from old links).

But if we go that route, we could also just put the new notebook in the main mellea repo (under docs/examples/notebooks/) and then retire this one.