Fix mypy type errors (117 errors across 23 files)

[mortentabor]

Summary

The CI runs mypy with strict = true and currently reports 63 errors across 12 source files. These are pre-existing type annotation issues — the Type Check job is set to continue-on-error: true so it doesn't block merging, but we should resolve them.

Error Breakdown by Category

no-any-return (10 errors)

Functions with return type annotations that return numpy/pandas values mypy can't verify.

File	Lines
models/ols.py	74, 88, 110
models/ar.py	102, 112
models/adl.py	119, 129
results/base.py	143
tests/bai_perron.py	452, 459, 632
visualization/params.py	145

Fix: Add explicit casts or adjust return type annotations.

arg-type — numpy dtype mismatches (17 errors)

Numpy signedinteger vs float64 dtype mismatches when appending to typed lists, plus ArrayLike | None vs ArrayLike issues.

File	Lines
models/ar.py	450, 455, 461, 610
models/adl.py	668, 674, 680, 790, 1032
rolling/ar.py	281, 286, 546, 551
rolling/adl.py	400, 406, 412, 443, 739, 745, 751, 782
tests/bai_perron.py	361
visualization/breaks.py	280

Fix: Use broader NDArray[np.floating[Any]] type annotations for lists, or cast arrays.

assignment — Figure type narrowing (6 errors)

ax.get_figure() returns Figure | SubFigure | None, but variables are typed as Figure.

File	Lines
visualization/breaks.py	132, 254
visualization/diagnostics.py	267, 391
visualization/params.py	568

Fix: Use assert isinstance(fig, Figure) or handle the None/SubFigure cases.

unused-ignore (5 errors)

Stale # type: ignore comments that no longer match the actual error codes.

File	Lines
visualization/breaks.py	132, 254
visualization/diagnostics.py	267, 391, 506

Fix: Remove or update the # type: ignore comments with correct error codes.

attr-defined (2 errors)

RegressionResultsBase has no attribute sigma.

File	Lines
visualization/diagnostics.py	381, 627

Fix: Add sigma to RegressionResultsBase or narrow the type to a subclass that has it.

return-value (3 errors)

Incompatible return types (e.g., returning None where ndarray expected, complex vs floating array).

File	Lines
models/ols.py	495
models/ar.py	78
models/adl.py	95

Fix: Adjust return type annotations or add runtime checks.

union-attr (1 error)

Accessing .get_figure() on Axes | Sequence[Axes].

File	Line
visualization/params.py	568

Fix: Narrow the type before calling .get_figure().

call-overload (1 error)

pd.DataFrame constructor type mismatch with dict of ndarrays.

File	Line
results/base.py	259

Fix: Adjust the dict value type or add a cast.

Files Affected

1. src/regimes/models/ols.py (5 errors)
2. src/regimes/models/ar.py (6 errors)
3. src/regimes/models/adl.py (7 errors)
4. src/regimes/rolling/ar.py (4 errors)
5. src/regimes/rolling/adl.py (8 errors)
6. src/regimes/tests/bai_perron.py (6 errors)
7. src/regimes/results/base.py (3 errors)
8. src/regimes/visualization/breaks.py (5 errors)
9. src/regimes/visualization/diagnostics.py (7 errors)
10. src/regimes/visualization/params.py (3 errors)

Suggested Approach

1. Fix unused-ignore comments first (quick wins, reduces noise)
2. Fix assignment / union-attr with type narrowing (assert isinstance(...))
3. Fix no-any-return with explicit casts
4. Fix arg-type dtype mismatches by broadening list type annotations
5. Fix attr-defined by adding sigma to the base class or narrowing types
6. Fix return-value issues with corrected annotations

CI Context

The Type Check job in CI uses continue-on-error: true so these errors are visible but non-blocking. Once resolved, we can remove continue-on-error to make mypy a hard gate.

[mortentabor]

Update (v0.3.0)

Error count increased from 63 errors across 12 files to 117 errors across 23 files after PR #5 (Markov switching implementation). The new errors come from the new source files:

• src/regimes/markov/models.py
• src/regimes/markov/restricted.py
• src/regimes/markov/results.py
• Other new v0.3.0 source files

New error categories introduced:

• misc (subclass Any) — MarkovSwitchingRestricted inheriting from a class mypy treats as Any
• Additional no-any-return, arg-type in Markov switching code

No change in approach — the Type Check job remains advisory (continue-on-error: true). This is a v0.5.0 task alongside documentation.

Updating title to reflect current count.