Preserve QRF feature order during prediction

[MaxGhenis]

Summary

• store fitted feature-column order for QRF and random-forest classifier learners and realign prediction frames before calling sklearn
• automatically keep numeric targets numeric when they become chained predictors inside zero-inflated per-regime base imputers
• pass explicit not_numeric_categorical overrides through zero-inflated per-regime base imputers too
• add focused regressions for reordered prediction frames, automatic chained numeric target preservation, and explicit override propagation

Tests

• uv run ruff check microimpute/models/qrf.py microimpute/models/zero_inflated.py
• uv run python -m pytest -q tests/test_models/test_zero_inflated.py::TestSequentialPredictorTyping tests/test_models/test_qrf.py::test_qrf_model_prediction_reorders_to_fitted_feature_order
• uv run python -m pytest -q tests/test_models/test_zero_inflated.py
• uv run python -m pytest -q tests/test_models/test_qrf.py

Notes

This fixes the Stage 05 MP/eCPS regime-aware resume failure where a later chained QRF prediction saw the same feature names in a different order than fit. It also prevents continuous prior imputed targets from being dummy-encoded inside nested zero-inflated base fits when they become predictors for later targets.

[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 9:43am