Tests

Two independent test paths.

Python suite — test_regression_macros.py

Checks every fit, predict, and evaluate against an equivalent scikit-learn model on the same fixed-seed data, so a failure means the macros disagree with a trusted reference (not just that a recorded number drifted). Covers all four families (including Tweedie across powers, negative binomial, multinomial softmax, offset/exposure, sample weights, k-fold cross-validation via cv_l2/cv_l1/ cv_power/cv_alpha, NB dispersion estimation via nbinom_dispersion, and two-stage grid refinement via reg_grid/cv_*_refine/nbinom_dispersion_refine, and Wald inference via *_summary — standard errors, z/t statistics, p-values and confidence intervals for all six single-outcome families plus multinomial (baseline-category Fisher information), robust HC0-HC3 and cluster-robust (sandwich) standard errors, prediction intervals (*_predict_ci, CI on the predicted mean), influence diagnostics (*_influence: leverage, Pearson/deviance residuals, studentized residuals, Cook's distance), and the IRLS solver (solver := 'irls'), all checked against an independent numpy/scipy reference, plus the pure-SQL norm_cdf/norm_ppf/t_cdf/t_ppf helpers vs SciPy), ridge/lasso/elastic-net with the documented sklearn equivalences (and KKT-optimality checks for L1 where sklearn has no reference), predict semantics, goodness-of-fit metrics (*_evaluate vs sklearn R²/AUC/log-loss/ deviance/d2_tweedie_score), NULL / constant-feature / type edge cases, and the error + reserved-name contract.

python -m venv .venv
.venv/Scripts/python -m pip install -r tests/requirements.txt   # Windows
# .venv/bin/python -m pip install -r tests/requirements.txt      # macOS/Linux
.venv/Scripts/python -m pytest tests/ -q

SQL smoke test — smoke.sql

No Python required — just the DuckDB CLI. Fits each family on deterministic inline data and aborts (non-zero exit) on the first failed check. Run from the repo root:

duckdb < tests/smoke.sql