better-regressions 0.5.0

Advanced regression methods with sklearn-like interface

Better Regressions

Advanced regression methods with an sklearn-like interface.

Current Features

• Linear:
  • Configurable regularization: Ridge with given alpha / BayesianRidge / ARD
  • "Better bias" option to properly regularize the intercept term
• Scaler:
  • Configurable preprocessing: Standard scaling (by second moment) / Quantile transformation with uniform/normal output / Power transformation
  • AutoScaler to automatically select the best scaling method based on validation split
• Smooth: Boosting-based regression using smooth functions for features
  • SuperSmoother: Adaptive-span smoother for arbitrary complex functions.
  • Angle: Bagging of piecewise-linear functions, it's less flexible but because of that it's more robust to overfitting.
• Soft: Mixture of regressors based on quantile classification
• Stabilize: Robust scaling & clipping transformation for features/targets
• AutoClassifier: Classification with automatic model selection (LogisticRegression or XGBoost, with auto depth selection)
• Adapter: Bins features and target, then trains a classifier. This way it can learn non-linear relationships and it also models the target distribution (not only its mean).
• EDA: Exploratory Data Analysis utilities
  • plot_distribution: Visualize sample distributions with fitted t-distribution parameters
  • plot_trend: Automatically detect and visualize relationships between variables + Pearson/Spearman correlation
    • For discrete features: Shows violin plots with distribution at each value
    • For continuous features: Fits trend lines with variance estimation and confidence intervals

Installation

pip install better-regressions

Basic Usage

from better_regressions import auto_angle, auto_linear, Linear, Scaler, AutoClassifier
from better_regressions.eda import plot_distribution, plot_trend
from sklearn.datasets import make_regression, make_moons
import numpy as np

X, y = make_regression(n_samples=100, n_features=5, noise=0.1)
model = auto_angle(n_breakpoints=2)
model.fit(X, y)
y_pred = model.predict(X)
print(repr(model))

# Classification example
dataset = make_moons(n_samples=200, noise=0.3)
Xc, yc = dataset
clf = AutoClassifier(depth="auto")
clf.fit(Xc, yc)
yc_pred = clf.predict(Xc)

# EDA example
plot_distribution(y, name="Target Distribution")
plot_trend(X[:, 0], y, name="Feature 0 vs Target")