nan_triangle 1D logic overhaul

[danielfong-act]

Fixes #692

1. Reimplemented is_pattern as a property in Triangle. Adds _pattern as an internal attribute.
2. Added the abstract properties is_pattern and is_ultimate to TriangleBase. TriangleBase is now an abstract class.
3. Changed nan_triangle to check explicitly for pattern/ultimate vectors, instead of checking for any 1D Triangles.

see this post. cum_to_incr() mishandles 2x2 triangles because it passes nan_triangle a one dimensional triangle (corresponding to the latter development period), and receives back a full 1D triangle of ones. This causes cum_to_incr() to ignore valuation information and populate the lower right corner of the triangle with a negative incremental value.

In addition to the test suite, I tested this PR with this snippet of code pulled from the issue discussion:

import chainladder as cl
import pandas as pd

df = pd.DataFrame(data={
    'origin': [2022, 2022, 2023],
    'development': [2022, 2023, 2023],
    'reported': [78000, 222000, 78000]}
)

tri_from_df = cl.Triangle(
    data=df,
    origin='origin',
    development='development',
    columns=['reported'],
    cumulative=True
)

tri_from_df

tri_from_df.cum_to_incr()

I also ensured that development estimators that set is_pattern (e.g. ClarkLDF) continue to function properly.

This PR will restrict nan_triangle to only return 1D triangles of ones for pattern and ultimate triangles. If I'm missing a case, please let me know.

---

Note

Medium Risk
Touches core triangle masking used across arithmetic and conversions, so it may subtly change results for pattern/ultimate workflows even though coverage includes a targeted regression test.

Overview
Fixes a cum_to_incr() regression on 2x2 triangles by changing nan_triangle to return an all-ones mask only for pattern vectors and ultimate vectors, rather than for any 1D triangle.

Promotes TriangleBase to an abstract base class by requiring is_pattern/is_ultimate properties, and reworks Triangle.is_pattern to use an internal _pattern attribute with a property/setter. Adds a regression test covering the 2x2 conversion case.

^{Reviewed by Cursor Bugbot for commit a25ea4c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.12%. Comparing base (08b9987) to head (a25ea4c).
⚠️ Report is 12 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #702      +/-   ##
==========================================
+ Coverage   85.08%   85.12%   +0.03%     
==========================================
  Files          85       85              
  Lines        4896     4909      +13     
  Branches      629      629              
==========================================
+ Hits         4166     4179      +13     
  Misses        521      521              
  Partials      209      209              

Flag	Coverage Δ
unittests	85.12% <100.00%> (+0.03%)	⬆️

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

☔ View full report in Codecov by Sentry.
📢 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.

[genedan]

@danielfong-act, thanks for going the extra mile and figuring out the abstract class features! I just have a few minor requests:

1. Add type hints to the getter and setter (-> bool)
2. Add a space after self, in line 492

[danielfong-act]

sure, cleaner code is always better

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 2efef59. Configure here.}

[cursor]

Reviewer requested -> bool but got -> np.bool

Low Severity

The is_ultimate property uses -> np.bool as the return type annotation. The PR reviewer explicitly requested -> bool type hints for the getter and setter. Additionally, np.bool was removed in numpy 1.24 and only restored in numpy 2.0, making it an unreliable alias. The function actually returns a Python bool (from int > 0), so the annotation is also technically inaccurate. This violates the reviewer's explicit request: "Add type hints to the getter and setter (-> bool)".

 

^{Reviewed by Cursor Bugbot for commit 2efef59. Configure here.}

[genedan]

Could you put in three more bool return type hints on these lines?

318
323
493

Also, please add a unit test for a 2x2 triangle. I think the one in #692 would be a good one to use.

[danielfong-act]

Sure, but does using numpy.bool_ (where appropriate) bother you? e.g. is_ultimate returns it since it uses numpy arrays.

[genedan]

numpy.bool_ works for me.

[danielfong-act]

I left the setter alone since python really seems to discourage having setters returning anything. I also don't think people expect a return in the context of a property assignment.

[genedan]

Closes #692. Good job!