Statsmodels

When you need actual statistical inference with p-values, confidence intervals, and diagnostic tests rather than just predictions, this is your tool. It handles the full spectrum from OLS and logistic regression to ARIMA time series and GLMs with proper hypothesis testing, AIC/BIC model comparison, and publication-ready summary tables. The library can feel academic compared to scikit-learn, but that's the point. You get residual diagnostics, heteroskedasticity tests, marginal effects, and all the econometric rigor that sklearn deliberately leaves out. One gotcha: you must manually add a constant with sm.add_constant() or your intercept will be zero, which trips up newcomers constantly.