Simplify solara code by Corvince · Pull Request #1786 · mesa/mesa

Corvince · 2023-08-31T19:43:39Z

Over the past weeks, I've had the opportunity to delve into the Solara frontend and experiment with different ideas. A big thank you to @rht for laying the foundation of the Solara frontend.

I'm experimenting with exciting concepts for the frontend and am eager to incorporate my React knowledge.

However, with this PR, my primary aim is to streamline the codebase before introducing any additional modifications.

At present, the frontend code is organized across three functions/classes:

JupyterContainer: This contains model control functions, grid plotting features, and logic to segregate model parameters into fixed and user-configurable ones.
MesaComponent: This establishes user-configurable controls, sets up the model, contains some model control functions, performs JupyterContainer manipulations, and holds the plotting logic.
JupyterViz: This Solara component essentially returns the MesaComponent and initiates the Container.

I've reorganized and simplified the structure. Now, JupyterViz stands as the primary component, managing user input (subject to further enhancement in an upcoming PR), initializing the model, and handling the plotting logic. The model control has been externalized to a separate component named ModelController.

While this PR might seem extensive, it's mainly due to the code repositioning to achieve a more intuitive flow. The actual modifications are minimal. Key changes include:

Removal of FigureMatplotlib dependencies. Presently, we're free from unnecessary re-renders, so this optimization isn't required.
Elimination of df from the container. Previously, I believe this was used to trigger re-renders. Now, we track the current step to achieve this. This approach also supports models without a data collector and is essential for upcoming enhancements to the model controller.
The split_model_params function has been made top-level and devoid of side effects.

I think that sums it up.

codecov · 2023-08-31T19:45:49Z

Codecov Report

Patch coverage: 11.36% and project coverage change: -0.44% ⚠️

Comparison is base (ea4b213) 80.02% compared to head (0c0b1f1) 79.58%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1786      +/-   ##
==========================================
- Coverage   80.02%   79.58%   -0.44%     
==========================================
  Files          15       15              
  Lines         881      877       -4     
  Branches      188      188              
==========================================
- Hits          705      698       -7     
- Misses        155      158       +3     
  Partials       21       21

Files Changed	Coverage Δ
mesa/experimental/jupyter_viz.py	`15.26% <11.36%> (-4.74%)`	⬇️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Corvince · 2023-08-31T19:55:52Z

Just to give some perspective on what "future improvements" from my side are already locally implemented (for testing, need more polishing)

Use more components => All the make_* functions can and should be converted to solara components (reuseability)
Run models using use_thread hook - this works on its own, but matplotlib is not thread-safe, so the line-plots crash (much simpler code, but needs some investigating)
Move all the input parameter logic into its own component (challenge how to trigger re-initializing of the model)
Some layout/styling improvements

rht · 2023-08-31T21:53:54Z

mesa/experimental/jupyter_viz.py

-                self.model_params_fixed[k] = v
+
+@solara.component
+def JupyterViz(


Yeah, this is simpler. The reason why I split into JupyterContainer, MesaComponent, and JupyterViz was the answer to these questions on Discord:

https://discord.com/channels/1106593685241614489/1106593686223069309/1123620981911597208

The question:

Is it possible to do just

model_params = { "N": { "type": "SliderInt", "value": 50, "label": "Number of agents:", "min": 10, "max": 100, "step": 1, }, "width": 10, "height": 10, } JupyterViz( BoltzmannWealthModel, model_params, measures=["Gini"], name="Money Model" )

The answer,

@rht you can also assign to page = MesaComponent(...) but they you would have to reverse your model, so JupyterViz shouldn't call MesaComponent, but the other way around

This was the reason for the MesaComponent and JupyterViz split. But in hindsight, the split wasn't necessary.

This was the reason for the JupyterContainer split: https://discord.com/channels/1106593685241614489/1106593686223069309/1123583246307954828.

Yes, the power of hindsight... but still quite impressive to make it work with all the necessary basic features! Solara uses quite a different paradigm, so not everything is clear from the beginning

rht · 2023-08-31T21:57:29Z

mesa/experimental/jupyter_viz.py

+        model.step()
+        set_current_step(model.schedule.steps)

    def do_play(self):


This still has self.

rht · 2023-08-31T21:57:51Z

mesa/experimental/jupyter_viz.py

+        while model.running:
            self.do_step()

    def threaded_do_play(self):


This shouldn't have self.

yes, removed

rht · 2023-08-31T22:07:53Z

Removal of FigureMatplotlib dependencies. Presently, we're free from unnecessary re-renders, so this optimization isn't required.
Elimination of df from the container. Previously, I believe this was used to trigger re-renders. Now, we track the current step to achieve this. This approach also supports models without a data collector and is essential for upcoming enhancements to the model controller.

In this PR, only solara.Markdown(md_text=f"**Step:** {current_step}") has a dependency to current_step, but not the FigureMatplotlib's. How are they notified of the step change then?

rht · 2023-08-31T22:08:59Z

mesa/experimental/jupyter_viz.py

 import mesa

-# Avoid interactive backend
-plt.switch_backend("agg")


This was added so that user code that does the pyplot interface (plt.plot, etc) can still work, even though not recommended.

ah, makes sense. I reverted the change

Corvince · 2023-09-01T07:40:47Z

Removal of FigureMatplotlib dependencies. Presently, we're free from unnecessary re-renders, so this optimization isn't required.
Elimination of df from the container. Previously, I believe this was used to trigger re-renders. Now, we track the current step to achieve this. This approach also supports models without a data collector and is essential for upcoming enhancements to the model controller.

In this PR, only solara.Markdown(md_text=f"**Step:** {current_step}") has a dependency to current_step, but not the FigureMatplotlib's. How are they notified of the step change then?

Ah, I think you got the dependency list of FigureMatplotlib backwards. In Solara, when the state changes (here: current_step), the component and every child component is re-rendered. So we don't need to notify FigureMatplotlib explicitly to update the plot. Only if we dont want it to redraw we can add a dependency list and only if the values of the dependency list change a new figure is drawn. But even without a dependency change it technically still re-renders, but uses the old (cached) image.

rht · 2023-09-01T07:48:01Z

I see, that makes sense. I will merge now.

rht · 2023-09-01T15:27:09Z

Forgot to mention that the visualization tutorial needs to be updated ASAP before the release (see https://mesa.readthedocs.io/en/stable/tutorials/visualization_tutorial.html). I can't do this within the next few days, and so, @ankitk50, if you have the time to do it, it would be great.

ankitk50 · 2023-09-21T08:27:15Z

Hi all, I started working on updating this tutorial. I get the following error while rendering the page:

Failed to load model class 'SheetModel' from module 'jupyter-vuetify'
makeError@http://localhost:8889/static/components/requirejs/require.js?v=d37b48bb2137faa0ab98157e240c084dd5b1b5e74911723aa1d1f04c928c2a03dedf922d049e4815f7e5a369faa2e6b6a1000aae958b7953b5cc60411154f593:168:26

Are we aware of this issue or I'm missing some config ?

rht · 2023-09-21T09:49:45Z

Try https://discourse.jupyter.org/t/ipysheet-in-tljh-failed-to-load-model-class-sheetmodel-from-module-ipysheet/18714/2.

rht · 2023-09-22T11:41:27Z

@ankitk50 I had already fixed the issue (#1792) because we wanted to release sooner. You should look for other issue instead. #1741 seems to be a low-hanging fruit.

ankitk50 · 2023-09-22T12:45:51Z

Thanks @rht, everything looked in order to me. Although I cannot get rid of the Jupyter notebook error. The error does not show on colab, so I use it for now.

rht · 2023-09-23T05:34:12Z

I don't recommend using Colab to develop for long term. You have to fix your local env problem soon.

Simplify solara code

9f5ae8c

rht reviewed Aug 31, 2023

View reviewed changes

rht mentioned this pull request Aug 31, 2023

Add docstring for jupyterviz make_user_input that documents supported inputs #1784

Merged

re-introduced backend switch and removed self

0c0b1f1

rht approved these changes Sep 1, 2023

View reviewed changes

rht merged commit fb81c1a into main Sep 1, 2023

rht deleted the solara-simplify branch September 1, 2023 07:49

rht mentioned this pull request Sep 7, 2023

visualization tutorial: Needs to be updated to reflect the simplified code in #1786 #1792

Closed

tpike3 added this to the Release 2.1.2 milestone Sep 19, 2023

rht mentioned this pull request Sep 30, 2023

Solara: Let users rewind sessions #1805

Closed

Uh oh!

Conversation

Corvince commented Aug 31, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov bot commented Aug 31, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Corvince commented Aug 31, 2023

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rht commented Aug 31, 2023

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Corvince commented Sep 1, 2023

Uh oh!

rht commented Sep 1, 2023

Uh oh!

rht commented Sep 1, 2023

Uh oh!

ankitk50 commented Sep 21, 2023

Uh oh!

rht commented Sep 21, 2023

Uh oh!

rht commented Sep 22, 2023

Uh oh!

ankitk50 commented Sep 22, 2023

Uh oh!

rht commented Sep 23, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Corvince commented Aug 31, 2023 •

edited

Loading

codecov bot commented Aug 31, 2023 •

edited

Loading