Unified Trustworthy ML Toolkit Ecosystem

Three mature Python libraries, one integrated story for trustworthy AI.

PyOD, PyGDebias, and PyGIP cover anomaly detection, fair graph learning, and graph intellectual property protection. This hub frames them as a shared toolkit layer with measurable adoption, ready-to-run APIs, and interactive pathways for new users.

Ask how to use it Compare toolkits
40M+ PyOD public download signal
60+ PyOD detectors
26 PyGDebias graph datasets
13 PyGDebias fair graph algorithms

Proposal-Aligned Aim

From separate packages to a single adoption narrative.

The page positions the three libraries as a coherent backend and frontend ecosystem: standardized installation, comparable metadata, unified entry points, and a guided interface that answers practical user questions without requiring a separate user-study section.

Adoption Signals

Popularity badges that make impact visible at a glance.

Outlier Detection

PyOD

GitHub

A battle-tested Python library for scalable anomaly and outlier detection, with classical, deep, LLM-assisted, and benchmark-backed model selection workflows.

PyOD GitHub stars PyOD GitHub forks PyOD PyPI downloads PyOD PyPI version
  • 60+ detection algorithms under a consistent API.
  • Adoption-heavy anchor for the overall ecosystem narrative.
  • Works as the anomaly-detection backend service in the platform.

Fair Graph Learning

PyGDebias

GitHub

A PyTorch Geometric-based benchmark and toolkit for graph debiasing, bundling datasets, fairness-aware algorithms, and unified evaluation flows.

PyGDebias GitHub stars PyGDebias GitHub forks PyGDebias PyPI downloads PyGDebias PyPI version
  • 26 graph datasets for fairness-oriented experiments.
  • 13 implemented algorithms spanning preprocessing, in-processing, and postprocessing.
  • Provides the graph-fairness service in the unified toolkit stack.

Graph IP Protection

PyGIP

GitHub

A modular framework for graph model intellectual property protection, including model extraction attacks, defense methods, datasets, and graph neural network models.

PyGIP GitHub stars PyGIP GitHub forks PyGIP PyPI downloads PyGIP PyPI version
  • Attack and defense modules for graph model IP risk analysis.
  • Experiment-ready structure for GNN models and datasets.
  • Adds the model-protection service to the integrated platform.

Unified Architecture

A platform layer that makes three libraries feel like one product.

01

Shared Registry

Expose each algorithm, dataset, metric, and tutorial through a searchable metadata layer.

02

Microservice APIs

Wrap PyOD, PyGDebias, and PyGIP behind consistent endpoints for run, compare, and export.

03

Visual Builder

Let users assemble detection, debiasing, and protection workflows without reading every API first.

04

LLM Assistant

Answer “which toolkit do I need?” and “how do I install it?” with grounded, copyable recipes.

Interactive Demo

Ask practical questions and get a focused answer.

Use the presets or type a phrase such as “fair graph learning”, “outlier detection”, “model extraction”, “install PyGDebias”, or “compare all”.

Choose a question to generate a short, presentation-ready answer.

Quick Start

Copyable install recipes for each audience.

pip install pyod

from pyod.models.iforest import IForest

clf = IForest()
clf.fit(X_train)
scores = clf.decision_scores_
pip install pygdebias

from pygdebias.datasets import Bail

dataset = Bail()
adj, features, labels, idx_train, idx_val, idx_test, sens = dataset.load_data()
pip install PyGIP

# Use PyGIP's modular attack/defense components to evaluate
# graph model IP risk under model extraction settings.

Source Links

PyOD GitHub PyOD PyPI PyGDebias GitHub PyGDebias PyPI PyGIP GitHub PyGIP PyPI