README.md

all_predict

Author: Santu Chall
Email: santuchal@gmail.com

Overview

all_predict is an advanced, production-ready alternative to LazyPredict, providing:

• 30+ regression models and 30+ classification models from scikit-learn, XGBoost, LightGBM, CatBoost, and more.
• Automatic evaluation and ranking of all models.
• Top-3 model hyperparameter tuning using GridSearchCV with robust, extended parameter grids for deeper optimization.
• Performance metrics, training/prediction time logging.
• Model saving/loading utilities.
• Comparison visualizations.

Installation

pip install all_predict

Quick Start

Regression Example

from sklearn.datasets import make_regression
from all_predict.regression import LazyRegressorPlus

X, y = make_regression(n_samples=500, n_features=10, noise=0.1, random_state=42)

reg = LazyRegressorPlus(verbose=True)
results, tuned = reg.fit(X, y)
print(results.head())
print(tuned)

Classification Example

from sklearn.datasets import make_classification
from all_predict.classification import LazyClassifierPlus

X, y = make_classification(n_samples=500, n_features=10, n_classes=2, random_state=42)

clf = LazyClassifierPlus(verbose=True)
results, tuned = clf.fit(X, y)
print(results.head())
print(tuned)

Robust Parameter Grids

The GridSearchCV now uses expanded hyperparameter grids tailored for each model type:

• Tree-based models: n_estimators, max_depth, min_samples_split, min_samples_leaf, max_features, bootstrap, learning rate (if applicable)
• Linear models: alpha, l1_ratio, regularization type, solver variations
• Boosting models: n_estimators, learning_rate, max_depth, colsample_bytree, subsample, regularization terms
• SVM: C, kernel, gamma, degree
• KNN: n_neighbors, weights, metric

These grids allow the tuner to explore broader, more robust parameter spaces for significantly better model performance.

Features

• Extended Model List: More models than LazyPredict.
• Top-3 GridSearch Tuning: Automated hyperparameter optimization with deep parameter grids.
• Persistence: Save and load best models.
• Visualization: Compare performance and timings.

Output Example

Regression

Tuned Models

Comparison with LazyPredict

License

MIT License.