cleands

cleands is a Python package for statistical modeling and data science that unifies regression, classification, clustering, distribution modeling, and dimension reduction under a single interface.
It aims to provide a full open-source alternative to packages like Stata, SAS, SPSS, and MATLAB, while remaining extensible and Pythonic.

Features

• Formula interface: Fit models directly from a Patsy-like formula, e.g. "y ~ x1 + x2 + x1:x2".
• Supervised learning: Linear regression, logistic regression, Poisson regression, k-nearest neighbors, recursive partitioning trees, ensembles (bagging, random forests), shrinkage methods (lasso, ridge, elastic net), etc.
• Classification: Logistic, multinomial, discriminant analysis, kNN classifiers, decision trees, random forests.
• Unsupervised learning: k-means clustering and other clustering algorithms (planned).
• Distributions: Parametric probability distributions with PDF, CDF, and likelihood-based inference.
• Utilities: Cross-validation, bootstrap, gradient descent, Newton’s method, and more.

See the TODO list for planned additions such as QDA, SVM, DBSCAN, Gaussian mixtures, quantile regression, splines, and LDV models (Tobit, truncated).

Installation

Install the latest release from PyPI:

pip install cleands

Documentation

Full API documentation, usage guides, and examples are available here:

• cleands Documentation Index

(Replace the links above with your actual GitHub Pages or Read the Docs URLs once deployed.)

Quick Start

Fit a linear regression model using formula notation:

import pandas as pd
from cleands.Prediction import LeastSquaresRegressor

# Example DataFrame
df = pd.DataFrame({
    "y": [1, 2, 3, 4, 5],
    "x1": [1, 2, 3, 4, 5],
    "x2": [2, 1, 2, 1, 2]
})

# Fit model with interaction term
model = LeastSquaresRegressor("y ~ x1 + x2 + x1:x2", data=df)

print(model.tidy)   # Coefficients with std errors, t-stats, and p-values
print(model.glance) # Model summary (R², AIC, BIC, etc.)

Logistic and Poisson regression use the same interface:

from cleands.Prediction import LogisticRegressor, PoissonRegressor

logit_model = LogisticRegressor("y ~ x1 + x2", data=df)
pois_model  = PoissonRegressor("y ~ x1 + x2", data=df)

k-means clustering (unsupervised):

from cleands.Clustering import kMeans

kmeans = kMeans("~x1+x2", data=df, k=2)
print(kmeans.groups)   # Cluster assignments
print(kmeans.means)    # Cluster centroids

Directory Structure

cleands/
│
├── base.py              # Abstract base classes (prediction, classification, clustering, distribution)
├── formula.py           # Formula parser for Patsy-like expressions
├── utils.py             # Utility functions (cross-validation, bootstrap, optimizers, etc.)
│
├── Prediction/          # Regression and supervised prediction models
│   ├── glm.py           # Least squares, logistic, Poisson regressors
│   ├── knn.py           # k-nearest neighbors regressors
│   ├── shrinkage.py     # Lasso, ridge, elastic net, etc.
│   ├── tree.py          # Recursive partitioning regressors
│   ├── ensemble.py      # Bagging, random forests
│
├── Classification/      # Classification models
│   ├── glm.py           # Logistic and multinomial classifiers
│   ├── knn.py           # kNN classifiers
│   ├── lda.py           # Linear discriminant analysis
│   ├── tree.py          # Recursive partitioning classifiers
│   ├── ensemble.py      # Bagging and random forest classifiers
│
├── Clustering/          # Unsupervised clustering
│   ├── kmeans.py        # k-means clustering
│
├── DimensionReduction/  # PCA, CCA, etc. (in progress)
├── Distribution/        # Probability distributions and tests

Roadmap

• Stepwise model selection
• Support for splines and GAMs
• More clustering methods (DBSCAN, Gaussian mixtures, hierarchical)
• Additional LDV models (Tobit, truncated regression)
• Expanded distribution families
• Neural networks and GLM trees

License

MIT License. See LICENSE for details.