Estimators

pyensmallen now exposes estimator classes for common supervised learning and econometrics workflows:

• LinearRegression
• LogisticRegression
• PoissonRegression

Basic usage

import pyensmallen as pye

model = pye.LogisticRegression(fit_intercept=True)
model.fit(X, y)

coef = model.coef_
intercept = model.intercept_
probs = model.predict_proba(X)

Fitted attributes

After calling fit, each estimator exposes:

• params_
• coef_
• intercept_
• objective_value_

For unregularized fits, the inference layer also exposes:

• covariance_
• std_errors_
• coef_std_errors_
• intercept_std_error_
• robust_covariance_
• robust_std_errors_
• robust_coef_std_errors_
• robust_intercept_std_error_

Inference helpers

Confidence intervals:

model.confidence_intervals()
model.confidence_intervals(covariance_type="robust")

Compact summary output:

summary = model.summary()
robust_summary = model.summary(covariance_type="robust")

summary() returns a numeric array with columns:

[coef, se, ci_lb, ci_ub]

Regularization

All three estimators currently accept:

• alpha
• l1_ratio

The intended use right now is ridge-like penalization. The L1 path is implemented with a smooth approximation to stay compatible with the current L-BFGS backend.

Inference is only exposed for unregularized fits (alpha == 0).

Covariance types

Supported inference modes:

• covariance_type="nonrobust": model-based covariance
• covariance_type="robust": sandwich covariance

For OLS, the robust path uses the score-based sandwich form. For logit and Poisson, the robust path uses QMLE-style sandwich inference.