feat: Display model seed & allow user to specify it in JupyterViz

[rht]

Fixes: #1812. The motivation can be found in that issue. It looks like this

The current behavior is that the seed is not changed even when you press the reset button. After all, it is now a user input param. However, if you force refresh the browser, the seed is still not changed. A more sustainable solution (so that user doesn't accidentally lose the seed) is to add a button to generate a new seed.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
Schelling	small	🔵 +0.8% [+0.5%, +1.1%]	🔵 +0.6% [+0.4%, +0.9%]
Schelling	large	🔵 +0.6% [-0.0%, +1.2%]	🔵 -3.0% [-5.2%, -1.0%]
WolfSheep	small	🔵 +0.7% [-0.6%, +1.9%]	🔵 +0.2% [-0.0%, +0.4%]
WolfSheep	large	🔵 +0.3% [-0.4%, +1.2%]	🔵 +0.5% [-1.8%, +3.4%]
BoidFlockers	small	🔵 +1.0% [+0.3%, +1.8%]	🔵 +1.9% [+1.2%, +2.7%]
BoidFlockers	large	🔵 +1.3% [+0.6%, +1.9%]	🔵 +2.1% [+1.6%, +2.5%]

[EwoutH]

Sounds useful, thanks! I will leave the formal review to Corvince.

One thought: Maybe we can have separate buttons for requesting a new seed and resetting the simulation to it's initial state.

[tpike3]

Can we explicitly pass in the random model seed so it is obvious to the user where the seed is coming form and what they are changing is they pass in their own?

[rht]

This is already an explicit argument. Specified by adding a "seed" entry to the model_params dictionary: https://github.com/projectmesa/mesa-examples/blob/main/examples/schelling_experimental/app.py#L21.

[rht]

I forgot to remove that line from the docstring.

[tpike3]

@rht I am sorry I don't mean to belabor this, but I feel like I am missing something here. So please bear with me, my understanding is below.

Scenario 1: User does not specify the seed so we want the UI to show the seed that was selected. Shouldn't the display of the random seed reference obj._seed

Scenario 2: User specifies the seed* then it would display the "seed" parameter or still reference obj._seed

Thank you for bearing with me, I know we have had issues with the seed in the past and just want to make doubly sure we have one solid pointer to the model seed.

Let me know what you think

*Apologies my earlier comment was very confusing. seed should come from model_params and be explicitly displayed not explicitly passed.

[rht]

Scenario 1: User does not specify the seed so we want the UI to show the seed that was selected. Shouldn't the display of the random seed reference obj._seed

The direction of the cause and effect should flow from the controls you see on the screen, and then the model properties. It can't be the obj._seed to have a value before the control.

If you notice, the seed value is displayed twice. The one below the information card is model._seed, displayed for debugging purpose. I will remove this once you agree to merge.

[rht]

Update:

• simplified the implementation by separating the seed from model parameters. The diff is smaller now.
• changed the play widget color to be blue
• added back the seed as an optional argument
• added a reseed button

To test this PR, you need to modify your model.py, __init__, e.g. for Schelling, to accept seed as an argument

    def __init__(self, width=20, height=20, density=0.8, minority_pc=0.2, homophily=3, seed=0):
        super().__init__(seed=seed)

[Corvince]

To test this PR, you need to modify your model.py, __init__, e.g. for Schelling, to accept seed as an argument

    def __init__(self, width=20, height=20, density=0.8, minority_pc=0.2, homophily=3, seed=0):
        super().__init__(seed=seed)

Hm, does that mean all the models need to accept a seed parameter? I don't think thats optimal

[rht]

Hm, does that mean all the models need to accept a seed parameter? I don't think thats optimal

make_model can be modified to have a try-except to catch the TypeError: Schelling.__init__() got an unexpected keyword argument 'seed' error to do a fallback to not showing the seed controls.

[rht]

Update: with 34a1853, no modification on existing models is necessary anymore. It works out of the box.

[tpike3]

LGTM - This is very cool, the model seed is one of those critical details that have burned many a modeler and I think making it explicit is an excellent idea.

To merge it is (1) fixing the tests and (2) removing the seed below the information card. Thanks @rht

[rht]

@tpike3 sorry for the slow response, still recovering. I fixed the test by using pytest-mock instead of unittest mock. And removed the seed below the information card. I request this PR to be rebased & merged instead of squashed & merged, because it has the 2 unrelated commits:

1. Change the play/stop button to blue
2. Code cleanup

[tpike3]

LGTM - I also like the change to the current_step as a parameter, at least for me much more intuitive.