[Jay-Naik2526]

Fixes #204.

Problem

A new AgentSet starts with an empty DataFrame that has no columns (see AgentSet.__init__), so it has no unique_id column until agents are added. AgentSetRegistry.add and AgentSetRegistry._recompute_ids read agentset["unique_id"] directly, so registering a set that has no agents yet raised polars.exceptions.ColumnNotFoundError: unable to find column "unique_id":

from mesa_frames import AgentSet, Model

class E(AgentSet):
    def step(self):
        pass

m = Model()
m.sets += E(m)  # ColumnNotFoundError

Change

• AgentSet.index returns an empty UInt64 series named unique_id when the column does not exist yet.
• AgentSetRegistry.add and _recompute_ids use agentset.index instead of reading the column directly.

AgentSet.pos is built on index, so it works for an empty set as well.

I kept the existing design (an empty AgentSet has no schema) because of the comment in __init__ about adding agents later. If you would rather give an empty AgentSet a unique_id schema instead, I am happy to rework it that way.

Tests

• tests/test_agentsetregistry.py::test_add_empty_agentset: registers an empty set, adds a populated set next to it, then removes the populated set while the empty one stays (this exercises _recompute_ids).
• tests/test_agentset.py::test_index_empty: index of a new AgentSet is an empty UInt64 series named unique_id.

Both new tests fail on main with the ColumnNotFoundError above and pass with this change. With MESA_FRAMES_RUNTIME_TYPECHECKING=1, the full suite gives 146 passed and 1 failed; the failure is tests/test_datacollector.py::TestDataCollector::test_postgress, which needs a running PostgreSQL database and fails the same way on main. ruff check and ruff format --check pass.

[Jay-Naik2526]

A note on the second commit: it comes from the pre-commit.ci autofix bot, not from this change. The pre-commit.ci check currently fails on main too (after the hook update in #205 it flags two list comprehensions in examples/boltzmann_wealth/performance_plot.py and examples/sugarscape_ig/performance_comparison.py), and the bot fixed them on this branch.

The actual change is only the first commit (4 files: agentset.py, agentsetregistry.py and their tests). I am happy to drop the bot commit if you would rather fix the examples in a separate PR.

[Jay-Naik2526]

Context on the red checks, in case it helps with review:

• The CI Checks failures are not caused by this change. main has been failing the same way in its scheduled runs since at least Oct 19. CI installs polars==2.1.0 (the repo's uv.lock pins 1.36.1), and with polars 2.x the shared test fixtures such as ExampleAgentSet / ExampleAgentSetA fail in setup with polars.exceptions.ShapeError: can't broadcast Series 'wealth' of length 4 to length 0. With polars 2.1.0 locally, 52 tests error at setup and 9 fail in tests/test_agentset.py and tests/test_agentsetregistry.py alone, all from those fixtures.
• The two tests added here avoid those fixtures and pass with both polars 1.36.1 and 2.1.0 (I updated the registry test in the last commit for that reason).
• I have not opened a separate issue for the polars 2.x breakage. If it would be useful, I can open one and look at making the fixtures compatible.