GroupBy: Add count and agg methods

[EwoutH]

Added two new methods to the GroupBy class:

• count: Returns the count of agents in each group.
• agg: Performs aggregation on a specific attribute across groups, applying a function like sum, min, max, or mean.

Usage examples

1. count():

   • Returns the count of agents in each group.
   • Usage Example:

     grouped_agents.count()

     Output:

     {
         "group_1": 10,  # 10 agents in group_1
         "group_2": 8,   # 8 agents in group_2
     }

2. agg(attr_name, func):

   • Aggregates the values of a specific attribute across each group using a provided function (e.g., sum, min, max, mean).
   • Usage Example:

     # Sum the wealth of all agents in each group
     grouped_agents.agg("wealth", sum)

     Output:

     {
         "group_1": 15000,  # Total wealth of agents in group_1
         "group_2": 12000,  # Total wealth of agents in group_2
     }

---

I'm a bit worried about our AgentSet and now GroupBy exploding our code base.

1. Is there any built-in functionality or libraries for this? It seems we're implementing very obvious behavior ourselves here.
2. Do we need to split into multiple files?
3. How do we keep this properly documented? Do we need guides for this?

(obviously tests are still needed)

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 -1.7% [-2.5%, -0.9%]	🔵 -0.4% [-0.5%, -0.3%]
BoltzmannWealth	large	🔵 +0.4% [-0.6%, +1.4%]	🔵 +0.0% [-2.9%, +3.3%]
Schelling	small	🔵 +0.2% [-0.1%, +0.4%]	🔵 -0.3% [-0.5%, -0.0%]
Schelling	large	🔵 -0.4% [-1.1%, +0.3%]	🔵 -2.4% [-3.7%, -1.1%]
WolfSheep	small	🔵 -2.0% [-2.9%, -1.2%]	🔵 -2.2% [-2.6%, -1.9%]
WolfSheep	large	🟢 -6.0% [-7.3%, -4.9%]	🟢 -12.7% [-14.6%, -10.9%]
BoidFlockers	small	🔵 -0.3% [-0.7%, +0.3%]	🔵 +1.2% [+0.6%, +1.8%]
BoidFlockers	large	🔵 +0.1% [-0.8%, +1.1%]	🔵 +1.2% [+0.5%, +1.8%]

[quaquel]

1. Not sure you need count. You can already do it via some_agentset.groupby(type).map(len). So, if you are worried about the number of lines of code in the code base, you could leave it out. I personally would not worry too much about the number of lines of code and am open to adding some shorthands for frequently used patterns (like getting the number of agents of a particular type or that meet a given criterion).
2. agg makes good sense and is a useful addition. I think I even mentioned adding it to groupby while adding it to agentset.

[adamamer20]

Although it can be expressed with agg, I am in favour of having a separate method for count. It just makes code more readable and easier to write.

[rht]

Shouldn't this be dict[Hashable, int] as well?

[EwoutH]

Sorry, missed this one.

Aren't all dicts by design required to have Hashable as a key? Seems a bit redundant to define that every time.

Like, the int part adds information. The Hashable doesn't (over Any).

[EwoutH]

Please "Unresolve conversation" if you have additional comments by the way.

[rht]

The Hashable adds information to the type checker (in an IDE) to constraint user code. Arguably, any dict[Any, *] should be dict[Hashable, *]. I don't have a citation on why this is better. ChatGPT (the fallback version after my quota of GPT-4o has run out) seems to agree.

[EwoutH]

Yeah that sounds about right.

A smart type checker should know you can’t use non-hashable types as dict keys.

[rht]

I tried Mypy:

from typing import Hashable, Any

a: dict[Hashable, int] = {[1]: 1}
a: dict[Any, int] = {[1]: 1}

It caught the first one, but doesn't consider the 2nd one to be a type error.

[EwoutH]

Added tests and cleaned up commits, ready for final review.

[quaquel]

Looks good