Phase 2: statscore-regression — OLS, GLM, Ridge, Lasso + Python

[armanrasta]

Summary

Linear and generalized linear models, regularized regression, diagnostics. Python bindings ship alongside.

Rust (statscore-regression)

• OLS via QR decomposition
• WLS, GLS
• GLM: logit, probit, Poisson, Binomial families + IRLS
• Ridge, Lasso (coordinate descent), Elastic Net
• Diagnostics: residuals, leverage, Cook's D, VIF

Python bindings (statscore.regression)

• fit() / predict() API similar to statsmodels/sklearn
• Coefficient tables, R², p-values
• Tests vs statsmodels to 1e-8 on 100 datasets

Done when

• OLS matches statsmodels
• Python module on PyPI (0.2.0)