Add validation to _data_setter for non-standard data node names

[drbenvincent]

Summary

PyMCModel._data_setter() hard-codes "X" and "y" as data node names. If a subclass uses different names, it silently misbehaves or throws a cryptic PyMC error. This PR adds validation that raises a clear ValueError with actionable guidance when expected data nodes are missing.

Fixes #663

Changes

• causalpy/pymc_models.py:

  • Added validation at the top of _data_setter() that checks "X" and "y" exist in self.named_vars before proceeding
  • Raises ValueError with guidance to override _data_setter() when nodes are missing
  • Updated class and method docstrings to document the override contract
  • No behavioral changes to existing logic — the pm.set_data call is unchanged

• causalpy/experiments/panel_regression.py:

  • Fixed pre-existing mypy type: ignore codes (attr-defined → union-attr)

• causalpy/tests/test_pymc_models.py:

  • Added NonStandardDataModel test subclass using renamed data nodes
  • Added TestDataSetterValidation with 2 tests:
    • test_mismatched_X_raises: verifies clear error when "X" is renamed
    • test_missing_y_raises: verifies clear error when only "y" is renamed

Design rationale

The issue proposed 5 possible directions. After examining all existing subclasses, I found that:

• All standard models (LinearRegression, WeightedSumFitter, SoftmaxWeightedSumFitter) use "X" and "y" — they work as-is
• Complex models (BayesianBasisExpansionTimeSeries) already override _data_setter entirely
• Models with fundamentally different data (StateSpaceTimeSeries, InstrumentalVariableRegression, PropensityScore) override predict or fit entirely

So the override contract already works for all existing subclasses. The missing piece was validation: a subclass that forgets to override _data_setter previously got a cryptic error. Now it gets a clear ValueError pointing to the solution.

I deliberately kept this minimal — no new class attributes or API surface — because the name-remapping-only case doesn't arise in any existing subclass and would add complexity for a speculative benefit.

Testing

• 2 new validation tests pass
• All 49 existing tests in test_pymc_models.py pass
• All 30 PyMC integration tests pass
• All 21 timeseries model coverage tests pass
• prek run --all-files clean

Checklist

• Backwards compatible (no behavioral changes)
• No new API surface
• Tests cover both X-missing and y-missing error paths
• Docstrings updated

[read-the-docs-community]

Documentation build overview

📚 causalpy | 🛠️ Build #32279314 | 📁 Comparing 08ee2c3 against latest (1cf3387)

  🔍 Preview build  

4 files changed

± 404.html
± _modules/causalpy/pymc_models.html
± api/generated/causalpy.pymc_models.PyMCModel.html
± _modules/causalpy/experiments/panel_regression.html

[codecov]

Codecov Report

❌ Patch coverage is 93.87755% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 94.44%. Comparing base (1cf3387) to head (08ee2c3).

Files with missing lines	Patch %	Lines
causalpy/tests/test_pymc_models.py	93.18%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #847      +/-   ##
==========================================
- Coverage   94.44%   94.44%   -0.01%     
==========================================
  Files          80       80              
  Lines       12707    12754      +47     
  Branches      770      773       +3     
==========================================
+ Hits        12001    12045      +44     
- Misses        498      500       +2     
- Partials      208      209       +1     

☔ 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.