solara_viz: Add borders around ContinuousSpace

[EwoutH]

For all intents and purposes, this should work. However, on my systems, I can't get it to visualise

It should looks something like this:

But in reality, on my system, it keeps looking like this:

I can add a title for some reason, but can't add borders or a background color.

Could it be that the background color and spines are overwritten later? @rht @quaquel @ankitk50, if one of you can explain or find the source of this madness, please.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
Schelling	small	🔵 -0.2% [-0.4%, -0.0%]	🔵 +0.2% [-0.0%, +0.4%]
Schelling	large	🔵 -0.7% [-1.2%, -0.2%]	🔵 -0.9% [-1.8%, -0.1%]
WolfSheep	small	🔵 -0.5% [-0.9%, -0.0%]	🔵 -0.6% [-0.8%, -0.5%]
WolfSheep	large	🔵 +3.6% [+0.6%, +6.2%]	🔵 +4.2% [+2.6%, +6.2%]
BoidFlockers	small	🔵 -0.2% [-0.7%, +0.3%]	🔵 +1.0% [+0.5%, +1.4%]
BoidFlockers	large	🔵 +0.4% [-0.1%, +0.9%]	🔵 -0.4% [-0.8%, +0.0%]

[rht]

This line is the cause:

mesa/mesa/experimental/components/matplotlib.py

Line 25 in b96db8f

space_ax.set_axis_off()

.

[EwoutH]

That's 45 minutes of my life I will never get back.

Now looks like this. Dotted line is for torus=True (when false it's solid). What does everyone think, is this better?

[EwoutH]

@quaquel @ankitk50 also please let me know what you think!

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
Schelling	small	🔵 -2.1% [-2.5%, -1.8%]	🔵 -0.4% [-0.5%, -0.2%]
Schelling	large	🔵 -1.9% [-2.2%, -1.7%]	🔵 -3.3% [-5.4%, -1.4%]
WolfSheep	small	🟢 -3.6% [-4.1%, -3.1%]	🔵 -0.1% [-0.3%, +0.0%]
WolfSheep	large	🟢 -6.4% [-9.2%, -3.7%]	🟢 -10.5% [-12.1%, -8.7%]
BoidFlockers	small	🔵 -3.0% [-3.4%, -2.7%]	🔵 +0.1% [-0.7%, +0.9%]
BoidFlockers	large	🔵 -2.8% [-3.4%, -2.3%]	🔵 +0.8% [+0.3%, +1.4%]

[EwoutH]

While I think these are good defaults, maybe we should allow passing the background color and turning the borders on or off.

[quaquel]

I started looking at the current code last week when I ran into the bug fixed this morning via #1984. I want to return to this at some point in the next few weeks. I see two options

1. Handle it like it is done in matplotlib and seaborne where you have various dicts that you can pass that control the behavior in detail.
2. Make it much easier to write your own plotting function that you pass to jupyterviz. Basically, specify the interface (so (keyword) arguments with which the function will be called and what the function should return. Ideally, you would model this on matplotlib.animation. So you have an init function and a function called to update the plot. I, however, don't know how easy that would be in Solara.

In the long run, I believe that the second option is much more flexible and easier to use. For example, in my exam model, I want to plot a background 2d KDE and have my agents on top of this. It is impossible to do this via (1) but easy to support via (2). In fact, I would only need to plot the KDE once in the init and only have to update the location of the agents while running the model.

[rht]

Now looks like this. Dotted line is for torus=True (when false it's solid). What does everyone think, is this better?

The dotted/solid line SGTM. The background grey color is debatable, but if you must provide a background color:

1. I'd prefer Seaborn's shade of grey. Because it is what everyone who use ggplot2-based libraries are used to.
2. Needs a way for users to easily change the background color.

[EwoutH]

Shall I do only the line for now in this PR? And then we can discuss the other stuff later?

[rht]

Shall I do only the line for now in this PR? And then we can discuss the other stuff later?

SGTM

[rht]

Handle it like it is done in matplotlib and seaborne where you have various dicts that you can pass that control the behavior in detail.

See also #1797. I checked that Seaborn doesn't have its own Axes.set API. And you'd have to reuse Matplotlib's Axes.set for a declarative axes configuration.

Make it much easier to write your own plotting function that you pass to jupyterviz. Basically, specify the interface (so (keyword) arguments with which the function will be called and what the function should return. Ideally, you would model this on matplotlib.animation. So you have an init function and a function called to update the plot. I, however, don't know how easy that would be in Solara.

See also #885. The solution in this issue is blocked from being merged because we are considering of migrating to Altair instead of Matplotlib. However, we should implement both Matplotlib and Altair viz in parallel: #1849 (reply in thread), and to make sure they have as similar API as possible.

[EwoutH]

@rht go ahead for final review and squash and merge.

@quaquel Do you want a 2.2.3 release, and if so, should this PR be included?

[quaquel]

So 2.2.3 would include the view limits and this? It would be nice for teaching and the exam if it all works. I will still make the exam to also work on older versions because students...

[EwoutH]

So 2.2.3 would include the view limits and this? It would be nice for teaching and the exam if it all works.

Ever got a release from an open-source release custom made for you, within a day? There you go!

I will still make the exam to also work on older versions because students...

Just add at the top of the notebook:

import mesa
if mesa.__version__ != "2.2.3"
    %pip install -U mesa==2.2.3