Implement experimental DataCollector API v3

[rht]

Given that the fall semester is starting soon, there is not enough time to build a comprehensive new data collection API. So this is what I have cooked this afternoon.
It addresses #1944 (comment):

1. You can collect based on agent types using custom groups
2. Also via custom groups
3. Single-use measure is not yet addressed
4. Not yet measured for performance
5. There is to_df in this PR, and this feature is actually used in visualization
6. TODO
7. TODO, but this doesn't affect the API
8. I prefer 2-level grouping instead of flat dictionary. This allows for easy export to DF
9. TODO

and #1944 (comment):

1. same as 2 above
2. already implemented
3. already implemented in to_df

TODO:

• error handling
• documentation
• testing
• examples

[rht]

For example, see mesa/mesa-examples#151.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
Schelling	small	🔵 -0.2% [-0.5%, +0.1%]	🔵 -0.7% [-0.8%, -0.5%]
Schelling	large	🔵 +0.3% [-0.6%, +1.1%]	🔵 -0.2% [-0.8%, +0.4%]
WolfSheep	small	🔵 -0.4% [-1.6%, +0.9%]	🔵 -0.7% [-1.0%, -0.4%]
WolfSheep	large	🔵 -0.9% [-1.6%, -0.3%]	🔵 -1.1% [-2.5%, -0.2%]
BoidFlockers	small	🔵 -1.3% [-1.8%, -0.7%]	🔵 -0.6% [-1.3%, +0.1%]
BoidFlockers	large	🔵 -1.2% [-1.8%, -0.8%]	🔵 +0.0% [-0.6%, +0.7%]

[rht]

This should be replaced with isinstance(group, AgentSet) or isinstance of Group.

[EwoutH]

Thanks for this effort! It doesn’t look too bad, could you give some example usages of more complicated usage scenarios?

[rht]

Which example do you want me to port? Is wolf-sheep (agents by type) sufficient?

[EwoutH]

I don't need a full model port, I would like some (more complex) API usage examples. Something in the form of:

• The future of data collection #1944 (comment)
• Multi-Agent data collection #348 (comment)

[rht]

@EwoutH this should be a sufficient illustration: mesa/mesa-examples#158. It has agents by type.

[quaquel]

This PR is stale, and we are at long last picking this up again.