Adds Altair plot component and some other fixes

[nissu99]

#2435

• - rename make_space_altair to something like make_altair_space_component
• - add make_altair_plot_component
• - Add support for all spaces. This means including the hexgrid and network transformation of the x and y coordinates.
• - Have a generic get_agent_data method

@quaquel @EwoutH I’ve added a new component, make_altair_plot_component. Could you kindly check if it aligns well with the project?

P.S working on space support.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 +1.3% [+0.4%, +2.2%]	🔵 -0.3% [-0.5%, -0.2%]
BoltzmannWealth	large	🔵 -1.2% [-1.8%, -0.6%]	🔵 -0.5% [-1.6%, +0.7%]
Schelling	small	🔵 -0.7% [-1.0%, -0.5%]	🔵 +0.8% [+0.6%, +0.9%]
Schelling	large	🔵 -0.7% [-1.1%, -0.3%]	🔵 -0.9% [-1.3%, -0.5%]
WolfSheep	small	🔵 +0.1% [-0.1%, +0.3%]	🔵 +1.1% [+0.9%, +1.2%]
WolfSheep	large	🔵 +0.9% [+0.3%, +1.5%]	🔵 +1.9% [+1.2%, +2.5%]
BoidFlockers	small	🔵 -0.7% [-1.3%, -0.2%]	🔵 +1.8% [+1.6%, +2.0%]
BoidFlockers	large	🔵 -0.1% [-0.5%, +0.4%]	🔵 +1.9% [+1.6%, +2.3%]

[EwoutH]

Thanks for the PR! A few of the automated checks broke, could you look into what caused that?

I will leave the review to others.

[tpike3]

@nissu99 also please look at #2643 and #2641 As you are @sanika-n are working in the same space

As it looks like you two are looking over the entire altair implementation I would also recommend looking at #2642 discussion

It is always good to collaborate and think together you may address some larger visualization challenges.

[nissu99]

@tpike3 When I run the tests locally, they don't fail, but they are failing here.

[tpike3]

@tpike3 When I run the tests locally, they don't fail, but they are failing here.

I am not sure why your tests are passing locally, however for tests to pass they need to updated. It is expecting None and now getting Chart, because you added post process.

__ ERROR collecting tests/test_components_matplotlib.py _____________
tests/test_components_matplotlib.py:[20](https://github.com/projectmesa/mesa/actions/runs/12984364297/job/36207123671?pr=2644#step:6:21): in <module>
    from mesa.visualization.mpl_space_drawing import (
mesa/visualization/__init__.py:13: in <module>
    from .components import make_plot_component, make_space_component
mesa/visualization/components/__init__.py:7: in <module>
    from .altair_components import (
mesa/visualization/components/altair_components.py:93: in <module>
    post_process: Callable[[alt.Chart], alt.Chart] | None = None,
E   AttributeError: 'NoneType' object has no attribute 'Chart'

[nissu99]

@tpike3 When I run the tests locally, they don't fail, but they are failing here.

I am not sure why your tests are passing locally, however for tests to pass they need to updated. It is expecting None and now getting Chart, because you added post process.

__ ERROR collecting tests/test_components_matplotlib.py _____________
tests/test_components_matplotlib.py:[20](https://github.com/projectmesa/mesa/actions/runs/12984364297/job/36207123671?pr=2644#step:6:21): in <module>
    from mesa.visualization.mpl_space_drawing import (
mesa/visualization/__init__.py:13: in <module>
    from .components import make_plot_component, make_space_component
mesa/visualization/components/__init__.py:7: in <module>
    from .altair_components import (
mesa/visualization/components/altair_components.py:93: in <module>
    post_process: Callable[[alt.Chart], alt.Chart] | None = None,
E   AttributeError: 'NoneType' object has no attribute 'Chart'

Thank You, the issue got resolved by adding altair to the toml file.

[nissu99]

@quaquel @tpike3 i was making a generic get_agent_data function ,should i remove all the old functions like _get_agent_data_new_discrete_space or keeping them will be right?

[quaquel]

Removing stuff that has become redundant is fine.

[nissu99]

@quaquel i have added all the stuff,can you review it?

[quaquel]

I'll try to review it over the weekend, but I am rather bussy at the moment. This is a part of the code base I am not intimately familiar with and his been a while since I used Altair, so reviewing will take more time than parts of the code base I know inside out.

[nissu99]

I have done a basic review, the grid drawing can wait, let's go step by step. Our first step here should be to finalize the agent data collecting logic, therefore why is the agent data collecting logic not made into a function like is done in matplotlib part? Also does this implementation cover all the grids (again compare with this)

stuff like vernoi grid has not been added ,will have to look into it to properly add it.

[Sahil-Chhoker]

@nissu99 I can see why you got confused, I think he meant the generic function should be called in the respective draw functions and hence agnet_portrayal was passed to the draw functions. Again if you look at the matplotlib side you will get it. I think the exact same method can be used to get the data, can you try importing the method from the matplotlib side and using it to test if that works?

[nissu99]

@nissu99 I can see why you got confused, I think he meant the generic function should be called in the respective draw functions and hence agnet_portrayal was passed to the draw functions. Again if you look at the matplotlib side you will get it. I think the exact same method can be used to get the data, can you try importing the method from the matplotlib side and using it to test if that works

I thought ,later he wants to make some universal function which can be reused.

[Sahil-Chhoker]

You don't have to worry about the new API, its still in discussion.

[nissu99]

You don't have to worry about the new API, its still in discussion.

i tried using the collect_agent_data function it does not give the desired results.

[Sahil-Chhoker]

How did you test it, can you please elaborate how is it not working?

[nissu99]

will ha

How did you test it, can you please elaborate how is it not working?

i used the internal examples ,somehow it shows no agent showing ,it is not able to extract the data ,will have to properly look into it to understand the issue well.

[nissu99]

[Sahil-Chhoker]

That's because you are just getting the agent data, no plotting them on the grid. Maybe check if the data extraction is working or not don't not rely on this visualization.

[nissu99]

@Sahil-Chhoker sorry for not getting back ,just got my college holidays started , i think the extraction is the fault here
when i printed the collect_agent_data returned values it was an empty array .

[Sahil-Chhoker]

Its working fine for me, can you test again. I don't think its a difficult task, you just replace the agent data collection logic with the collect_agent_data function call inside the grid draw logic.

[nissu99]

Its working fine for me, can you test again. I don't think its a difficult task, you just replace the agent data collection logic with the collect_agent_data function call inside the grid draw logic.

i get that , it is not working for OrthogonalMooreGrid somehow,for others it works fine.

[Sahil-Chhoker]

It's working for OrthogonalMooreGrid as well. Please review your code carefully before commenting, as I may not have the capacity to review it multiple times.

[nissu99]

It worked ,i made a dumb mistake ,should i make a commit for changing the extraction logic?

[Sahil-Chhoker]

Yeah, refactor the agent placement logic inline with the extraction logic.

[Sahil-Chhoker]

I encountered several issues while reviewing this PR. First, the previous review comments have not been addressed. Additionally, you're not using the new grids in mesa.discrete_space, among other issues.

Due to the excessive length of this PR, it's difficult for me to review it entirely. I strongly recommend breaking it down into smaller, atomic PRs that address specific issues.

A better approach could be:

1. Implement rectangular grids with agent placement logic.
2. Implement hex grids with agent placement logic.
3. Implement continuous space with agent placement logic.
4. Implement Voronoi grids with agent placement logic.
5. Implement network grids with agent placement logic.
6. Address any remaining grid implementations.
7. Add the plotting function.

This will make the review process smoother and ensure that changes are properly implemented step by step.

[nissu99]

I will change all the experimental to discrete_space ,I agree the changes have increased and i understand it is becoming very hard to review , is the extraction logic fine ?

[Sahil-Chhoker]

I have only tested it once, so it may be a good starting point. And again I strongly recommend you to break this PR into smaller once.

[Corvince]

And again I strongly recommend you to break this PR into smaller once.

Maybe to clarify what this means: Just open a series of new PRs, one after another. So start with some minimal changes as a single PR, which includes the groundwork for the rest. Once this PR is merged, open the next PR with changes to some other function/grid/whatever. And so on. Its more tedious this way, but overall it will speed things up. Right now the PR is too big to properly review, because it requires code review, which is already almost too much, but also manually testing everything.

You can also just leave this PR open for now as a reference.

[Sahil-Chhoker]

Thanks for the clarification @Corvince!

[nissu99]

And again I strongly recommend you to break this PR into smaller once.

Maybe to clarify what this means: Just open a series of new PRs, one after another. So start with some minimal changes as a single PR, which includes the groundwork for the rest. Once this PR is merged, open the next PR with changes to some other function/grid/whatever. And so on. Its more tedious this way, but overall it will speed things up. Right now the PR is too big to properly review, because it requires code review, which is already almost too much, but also manually testing everything.

You can also just leave this PR open for now as a reference.

Thank You for clarity ,will be making individual PR's .

[quaquel]

closing as stale