Remove schedulers from benchmark models.

[quaquel]

this removes the schedulers from the benchmark models (in one case it was in the init but not used in step anyway).

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔴 +20.3% [+18.9%, +21.7%]	🔴 +23.7% [+23.4%, +23.9%]
BoltzmannWealth	large	🔴 +28.6% [+28.1%, +29.1%]	🔴 +22.0% [+21.1%, +22.8%]
Schelling	small	🟢 -8.0% [-8.4%, -7.6%]	🔵 -1.0% [-1.2%, -0.8%]
Schelling	large	🟢 -9.0% [-9.4%, -8.5%]	🔵 -1.7% [-2.2%, -1.2%]
WolfSheep	small	🔵 -0.7% [-0.9%, -0.5%]	🔵 -0.0% [-0.2%, +0.1%]
WolfSheep	large	🔵 -2.9% [-3.8%, -2.0%]	🔵 -2.2% [-4.8%, -0.2%]
BoidFlockers	small	🟢 -8.0% [-8.8%, -7.4%]	🔵 +0.9% [+0.2%, +1.6%]
BoidFlockers	large	🟢 -8.4% [-8.9%, -8.0%]	🔵 -0.3% [-1.2%, +0.7%]

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔴 +23.1% [+21.9%, +24.3%]	🔴 +24.5% [+24.3%, +24.7%]
BoltzmannWealth	large	🔴 +29.4% [+29.0%, +29.7%]	🔴 +23.9% [+23.3%, +24.7%]
Schelling	small	🟢 -8.0% [-8.4%, -7.6%]	🔵 -1.2% [-1.4%, -1.0%]
Schelling	large	🟢 -8.7% [-8.9%, -8.5%]	🔵 -1.1% [-2.1%, -0.2%]
WolfSheep	small	🔵 +0.1% [-0.1%, +0.2%]	🔵 +0.0% [-0.2%, +0.2%]
WolfSheep	large	🔵 -1.1% [-1.4%, -0.7%]	🔵 +0.3% [-0.2%, +0.7%]
BoidFlockers	small	🟢 -8.0% [-8.5%, -7.4%]	🔵 -0.4% [-1.0%, +0.2%]
BoidFlockers	large	🟢 -8.3% [-8.7%, -7.9%]	🔵 -1.7% [-2.1%, -1.1%]

[EwoutH]

I don’t understand how BoltzmannWealth gets slower while doing less (being already on shuffle_do()).

Probably just a fluke.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔴 +28.7% [+27.0%, +30.2%]	🔴 +25.4% [+25.2%, +25.6%]
BoltzmannWealth	large	🔴 +30.0% [+29.6%, +30.3%]	🔴 +33.5% [+32.7%, +34.3%]
Schelling	small	🟢 -8.0% [-8.3%, -7.6%]	🔵 -0.1% [-0.3%, +0.1%]
Schelling	large	🟢 -7.6% [-8.3%, -6.9%]	🔴 +9.7% [+7.8%, +11.8%]
WolfSheep	small	🔵 +0.5% [+0.2%, +0.9%]	🔵 +1.8% [+1.6%, +2.0%]
WolfSheep	large	🔵 +2.1% [+0.1%, +4.1%]	🔵 +7.4% [+2.6%, +12.3%]
BoidFlockers	small	🟢 -7.8% [-8.4%, -7.1%]	🔵 +1.9% [+1.0%, +2.7%]
BoidFlockers	large	🟢 -9.0% [-9.2%, -8.7%]	🔵 +0.8% [+0.4%, +1.3%]

[quaquel]

I think it has to do with the data collection in the Boltzman wealth model. I need to dig deeper to confirm.

[quaquel]

My hunch is correct. The old boltzman wealth model contained a bug. It used a schedule but no agents were added to it. The datacollecter then collected data from the agents in the schedule (which were None). So, the new benchmark is correct.

Also, this means that the deprecation of schedules should result in an update to the datacollector. In fact, I think the entire agent_collector can be deprecated in favor of the new agent type stuff.

[EwoutH]

Yeah, that's correct, you can also see it in #2300. These two are completely equivalent:

 self.datacollector = DataCollector(
     agent_reporters={"energy": "energy"},
     agenttype_reporters={
         Agent: {"energy": "energy"}
     }

We will redesign this whole thing any way. Let's leave the API alone for now.

Implementation wise, these parts could be updated as soon as the schedulers are actually removed.

mesa/mesa/datacollection.py

Lines 231 to 237 in e6874ad

	agent_records = map(
	get_reports,
	model.schedule.agents
	if hasattr(model, "schedule") and model.schedule is not None
	else model.agents,
	)
	return agent_records

In between we could add a warning. But I don't think that's easy or worth it.

This PR is good to go!