Util weight

[henrydingliu]

Summary of Changes

Related GitHub Issue(s)

Additional Context for Reviewers

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

---

Note

Medium Risk
New public API and ~500 lines of weight logic that must stay consistent with Development; wrong ordering or preserve behavior would change LDFs once integrated, though production Development paths are unchanged in this diff.

Overview
Introduces TriangleWeight, a new sklearn-style utility that builds a weight triangle from the same exclusion knobs as Development (n_periods, drop, drop_valuation, drop_above/drop_below, drop_high/drop_low, preserve). It is exported from chainladder.utils and the top-level chainladder package, with EXPECTED_PUBLIC_API updated accordingly.

The implementation lives in chainladder/utils/triangle_weight.py and documents a different drop order than Development: threshold drops (drop_above/drop_below) run before rank-based drop_high/drop_low (the docstring notes steps 4 and 5 are reversed relative to Development). fit/transform set w_ on triangles and can compose with existing w_ weights.

Tests in test_development.py are expanded with _FutureDevelopment, a TriangleWeight subclass that recomputes LDFs via weighted regression, and assertions that its LDFs match Development across drop scenarios—exercising the new utility without wiring it into Development yet.

^{Reviewed by Cursor Bugbot for commit 60cabb2. Bugbot is set up for automated code reviews on this repo. Configure here.}

[codecov]

Codecov Report

❌ Patch coverage is 90.00000% with 16 lines in your changes missing coverage. Please review.
✅ Project coverage is 88.53%. Comparing base (1dd5f1d) to head (60cabb2).

Files with missing lines	Patch %	Lines
chainladder/utils/triangle_weight.py	89.20%	9 Missing and 6 partials ⚠️
chainladder/development/tests/test_development.py	95.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@                   Coverage Diff                    @@
##           #932_triangle_weight     #940      +/-   ##
========================================================
+ Coverage                 88.30%   88.53%   +0.22%     
========================================================
  Files                        88       89       +1     
  Lines                      5029     5189     +160     
  Branches                    642      658      +16     
========================================================
+ Hits                       4441     4594     +153     
+ Misses                      443      442       -1     
- Partials                    145      153       +8     

Flag	Coverage Δ
unittests	88.53% <90.00%> (+0.22%)	⬆️

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.

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

There are 3 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit ca2f50d. Configure here.}

[cursor]

Prior weights use wrong attribute

Medium Severity

When prior weights exist, TriangleWeight.fit multiplies the input triangle by X.w_ and passes that triangle into _set_weight_func. Development.fit instead checks w_v2_ and combines prior weights on obj.age_to_age. After Development, w_ and w_v2_ can differ, so chained weighting no longer matches development’s v2 path.

 

^{Reviewed by Cursor Bugbot for commit ca2f50d. Configure here.}

[cursor]

Ratio drops need age-to-age

Medium Severity

TriangleWeight.fit feeds the caller’s triangle straight into _set_weight_func, so _drop_x_func and _drop_n_func rank and threshold X.values as given. Those steps expect age-to-age link ratios; Development always builds v2 weights from obj.age_to_age. The same cumulative triangle passed to Development.fit can get incorrect exclusions here.

 

^{Reviewed by Cursor Bugbot for commit ca2f50d. Configure here.}