logitPY

LogitPY is a Python library that implements logistic regression and related generalized linear modeling tools. It provides estimation, inference, and model diagnostics for binary and multiclass outcomes, with an emphasis on a clean API, numerical stability, and compatibility with the Python data ecosystem (NumPy, SciPy, and pandas).

Key features include binary logistic regression and multinomial logit for multi-class problems via a one-vs-rest approach, regularization with L1, L2, and elastic net, and support for intercept and offset terms. The package offers statistical inference capabilities such as standard errors, Wald tests, and confidence intervals for coefficients. Optimization options include iteratively reweighted least squares (IRLS), Newton-Raphson, and gradient-based solvers with automatic convergence checks. Additional utilities cover feature scaling, basic missing-value handling, and cross-validation workflows.

API and usage follow a scikit-learn-inspired style. Users construct a model with y and X, call fit, and then obtain predictions or probabilities for new data. Example usage: model = Logit(y, X); model.fit(); probs = model.predict_proba(X_new); preds = model.predict(X_new). The library also provides methods such as get_params, set_params, and score to facilitate integration with pipelines and evaluation routines. Input data accepted includes NumPy arrays and SciPy sparse matrices, with compatibility considerations for common preprocessing steps.

History and licensing indicate an open-source project aimed at providing a lightweight alternative to heavier statistical packages. Documentation, tutorials, and community guidelines accompany releases, along with contribution and issue-tracking workflows. See also logistic regression, generalized linear models, and probabilistic programming tools.