Welcome to Nonconform's Documentation!

Nonconform is a Python library for conformal anomaly detection that provides uncertainty quantification for PyOD detectors. It implements statistically rigorous anomaly detection with p-values and error control metrics like False Discovery Rate (FDR).

Quick Links

• Installation - Get started with Nonconform
• Quick Start - Basic usage examples
• User Guide - Comprehensive documentation
• Examples - Practical examples and tutorials
• API Reference - Complete API documentation
• Contributing - How to contribute to the project

Key Features

• Conformal Inference: Distribution-free uncertainty quantification
• PyOD Integration: Works with popular anomaly detection library
• Multiple Strategies: Split, Bootstrap, Cross-validation, Jackknife+
• FDR Control: False Discovery Rate control for multiple testing
• Weighted Conformal: Handle non-exchangeable data

Getting Started

Install Nonconform with pip:

pip install nonconform

Basic usage:

from nonconform.estimation import ConformalDetector
from nonconform.strategy import Split
from pyod.models import IForest

# Create conformal detector
detector = ConformalDetector(IForest(), Split())

# Fit and predict with p-values
detector.fit(X_train)
p_values = detector.predict(X_test)