Chain numeric zero-inflated imputations

[MaxGhenis]

Summary

• make ZeroInflatedImputer fit numeric targets in declared order with earlier numeric targets added as predictors for later numeric targets
• predict with an augmented recipient frame so later targets receive prior imputed values
• add a regression test proving the second numeric target gets and uses the first target prediction

Tests

• uv run --extra dev python -m pytest -q tests/test_models/test_zero_inflated.py
• uv run --extra dev python -m pytest -q tests/test_models/test_qrf.py -k sequential
• uv run --extra dev python -m pytest -q tests/test_models/test_zero_inflated.py tests/test_models/test_qrf.py -k 'zero_inflated or sequential'
• uv run --extra dev towncrier check --compare-with origin/main
• uv run --extra dev ruff check microimpute/models/zero_inflated.py tests/test_models/test_zero_inflated.py
• uv run --extra dev ruff format microimpute/models/zero_inflated.py tests/test_models/test_zero_inflated.py --check

Context

Microplex was calling the zero-inflated wrapper with income/tax blocks, but numeric targets still lost QRF's sequential target chaining because the wrapper fit each numeric base imputer with only [variable]. This PR keeps the regime-aware gates while preserving numeric chained-equation behavior.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
microimpute-dashboard	Ready	Preview, Comment	Jun 6, 2026 6:25am

[MaxGhenis]

Closing as superseded by #193, which includes the same numeric zero-inflated chaining fix plus an explicit sequential flag and stronger joint-correlation regression. I pushed a formatter-only fix to #193 and will monitor that CI path instead.