[FIX] CapeCod.predict discards the fitted apriori when the prediction data has exactly one extra index level

[ppcvote]

Summary of Changes

CapeCod.predict() regroups the prediction data up to the grain the model was fit at, and decides whether to do so by counting the index levels sample_weight carries that apriori_ does not. The count required more than one, so the ordinary case of a single extra level fell through to the ungrouped branch and recomputed the apriori from the prediction data instead of using the fitted one.

import chainladder as cl

clrd = cl.load_sample("clrd")
tri, prem = clrd["CumPaidLoss"], clrd["EarnedPremDIR"].latest_diagonal

model = cl.CapeCod().fit(tri.groupby("LOB").sum(),
                         sample_weight=prem.groupby("LOB").sum())
model.predict(tri, sample_weight=prem)

The fitted comauto apriori is 0.5689995797. Before this change predict() returned 1.2516635774, and all 775 rows differed from the value fitted for their LOB. After it, every row carries its fitted apriori and expectation_ is exactly premium times that apriori across all 775 rows.

test_capecod_predict2 already covers this code path, but it uses prism, whose triangle carries five index levels more than the fitted model, so it stays on the grouped branch. The added test uses clrd, which carries exactly one.

Related GitHub Issue(s)

Fixes #1265. The API question that thread raised is split out into #1274.

Additional Context for Reviewers

• pytest chainladder passes: 1130 passed, 7 skipped.
• ruff check and ruff format --check under the CI flags report the same eight findings on these two files before and after the change, so the added lines introduce none of their own.
• The new test fails on the unpatched line and passes on the patched one.
• The predict() docstring example (ukmotor) is unaffected: it fits and predicts at the same grain, so the level difference is empty and neither branch changes. Its documented output still reproduces exactly.

Declarations

I am adhering to the standards in the Governing Doc. I am a human contributor, not a bot, and I opened this because @henrydingliu asked for the > 0 fix in #1265.

AI disclosure, per the AI Usage Policy: I used Claude Code on this. It reproduced the bug against clrd, ran the before and after comparison, and drafted the regression test. The one-character fix itself is the one @henrydingliu and I settled on in #1265 before any code was written. I reviewed the diff and the test myself, and the suite was run locally on my machine.

Checklist

• I passed tests locally for both code (uv run pytest) and documentation changes (uv run --directory docs jb build . --builder=custom --custom-builder=doctest)

  pytest locally: 1130 passed, 7 skipped. There are no documentation changes in this PR, so I did not run the docs doctest build. The only docstring the change could reach is the CapeCod.predict ukmotor example, and I ran that directly: it still reproduces its documented output exactly. CI's Doctest job passes on this branch.

[github-actions]

Pyright Type Completeness

View the full pyright --verifytypes output for this commit

Project (full chainladder package, at this PR's head): 14.9% of exported symbols fully typed (204 / 1370)

	Known	Ambiguous	Unknown	Total
Project (head)	204	111	1055	1370

Other symbols referenced but not exported by chainladder: 13

	Known	Ambiguous	Unknown	Total
Other (head)	3	1	9	13

Symbols without documentation:

• Functions without docstring: 325
• Functions without default param: 0
• Classes without docstring: 10

Patch (exported symbols added or changed by this PR): 0.0% fully typed (0 / 1)

	Known	Ambiguous	Unknown	Total
Patch	0	0	1	1

Patch symbol details

Symbol	Status	Change
chainladder.methods.tests.test_capecod.test_capecod_predict_one_extra_index_level	❌ unknown	new

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.69%. Comparing base (acd0bdb) to head (f036ba7).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1275      +/-   ##
==========================================
- Coverage   91.69%   91.69%   -0.01%     
==========================================
  Files          96       96              
  Lines        5457     5456       -1     
  Branches      702      702              
==========================================
- Hits         5004     5003       -1     
  Misses        328      328              
  Partials      125      125              

Flag	Coverage Δ
unittests	91.69% <100.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[henrydingliu]

looks good! would you mind picking up the couple ruff linter fixes?

[ppcvote]

Done. ruff check --force-exclude --config lint.per-file-ignores={} now passes on both files: the numpy import in capecod.py was unused, the rest was whitespace in existing tests. I also dropped the two per-file-ignores entries for these files, since the table's comment says to remove them as files are cleaned up and both are dead now. Rebased onto current main.

I left ruff format alone. It touches 40 of the 107 lines in test_capecod.py and grows it to 132, and six of the eight files under chainladder/methods are unformatted on main, so it looked like a separate decision rather than something to slip into this PR. Say the word if you would rather have it here.

[henrydingliu]

thanks again!