xplia 1.0.1

XPLIA: The Ultimate State-of-the-Art AI Explainability Library

XPLIA: The Ultimate State-of-the-Art AI Explainability Library

Production-grade explainability for trustworthy AI systems

Features • Installation • Quick Start • Documentation • Examples • Contributing

---

🎯 What is XPLIA?

XPLIA (eXplainable AI Library) is the most comprehensive, production-ready explainability framework for AI/ML systems. Built for enterprise deployments, XPLIA provides:

• 8+ Explanation Methods: SHAP, LIME, Counterfactuals, Gradients, Anchors, and more
• Regulatory Compliance: GDPR, EU AI Act, HIPAA compliance tools built-in
• Trust Evaluation: Uncertainty quantification, fairwashing detection, confidence reports
• Framework Agnostic: Works with scikit-learn, TensorFlow, PyTorch, XGBoost, LightGBM, CatBoost
• Production Ready: REST API, Docker, Kubernetes, MLflow/W&B integration
• Enterprise Grade: Audit trails, compliance reports, multi-audience explanations

Perfect for: Financial services, Healthcare, Legal tech, Government, Any regulated industry

---

✨ Features

🔍 Comprehensive Explainability Methods

Local Explanations SHAP - Best for tree models LIME - Model-agnostic, fast Gradients - For neural networks Counterfactuals - "What-if" scenarios Anchors - Rule-based explanations

Global Explanations Feature Importance - Model-wide patterns Partial Dependence - Feature effects Unified Explainer - Combines multiple methods Attention Maps - For transformers Model Summary - Complete overview

🏛️ Regulatory Compliance (Industry-First!)

from xplia.compliance import GDPRCompliance, AIActCompliance

# GDPR Right to Explanation
gdpr = GDPRCompliance(model)
dpia_report = gdpr.generate_dpia()  # PDF report ready for auditors

# EU AI Act Risk Assessment
ai_act = AIActCompliance(model, usage='credit_scoring')
risk = ai_act.assess_risk_category()  # Returns 'HIGH', 'MEDIUM', etc.
compliance_report = ai_act.generate_report()  # Full compliance documentation

Supported Regulations:

• ✅ GDPR - Right to explanation, DPIA generation
• ✅ EU AI Act - Risk assessment, documentation
• ✅ HIPAA - Healthcare compliance
• 🔜 SOC 2, ISO 27001 (Coming in v1.1)

🛡️ Trust & Confidence Evaluation

Uncertainty Quantification

Measure prediction confidence with 6 types of uncertainty:

from xplia.explainers.trust import UncertaintyQuantifier

uq = UncertaintyQuantifier(model, explainer)
uncertainty = uq.quantify(X_test)

print(f"Epistemic uncertainty: {uncertainty.epistemic_uncertainty}")
print(f"Aleatoric uncertainty: {uncertainty.aleatoric_uncertainty}")

Fairwashing Detection (Unique to XPLIA!)

Detect deceptive explanations that hide bias:

from xplia.explainers.trust import FairwashingDetector

detector = FairwashingDetector(model, explainer)
result = detector.detect(X_test, y_test)

if result.detected:
    print(f"⚠️ Fairwashing detected! Types: {result.fairwashing_types}")
    print(f"Severity: {result.severity}")

Detection Types:

• Feature masking
• Importance shift
• Bias hiding
• Cherry picking
• Threshold manipulation

🎨 Advanced Visualizations

from xplia.visualizations import ChartGenerator

generator = ChartGenerator()

# 12+ chart types
generator.create_chart(
    chart_type='waterfall',  # bar, line, pie, heatmap, radar, sankey, etc.
    data=explanation.feature_importance,
    title='Feature Importance',
    theme='dark',  # light, dark, corporate
    export='report.html'  # html, png, pdf, svg
)

🌐 Multi-Audience Adaptation

Automatic explanation adaptation for different audiences:

from xplia.explainers.calibration import AudienceAdapter

adapter = AudienceAdapter()

# Technical explanation for data scientists
tech_exp = adapter.adapt(explanation, audience='expert')

# Business-friendly explanation for executives
business_exp = adapter.adapt(explanation, audience='basic')

# Public explanation for end users
public_exp = adapter.adapt(explanation, audience='novice')

Audience Levels:

• 👨‍💼 Novice - General public
• 📊 Basic - Business stakeholders
• 🔬 Intermediate - Analysts
• 🎓 Advanced - Data scientists
• 👨‍🔬 Expert - ML researchers

---

🚀 Installation

Basic Installation (Lightweight, ~200MB)

pip install xplia

Includes: Core framework, basic visualizations, scikit-learn support

Full Installation (Recommended, ~2GB)

pip install xplia[full]

Includes everything: All XAI methods, deep learning, boosting, visualizations, APIs, ML Ops

Custom Installation (Choose what you need)

# XAI methods only
pip install xplia[xai]

# Deep learning support
pip install xplia[pytorch]  # or tensorflow

# Gradient boosting
pip install xplia[boosting]

# Advanced visualizations
pip install xplia[viz]

# API integrations
pip install xplia[api]

# ML Ops (MLflow, W&B)
pip install xplia[mlops]

# Development tools
pip install xplia[dev]

# Combine multiple
pip install xplia[xai,pytorch,viz,mlops]

From Source (Latest Development)

git clone https://github.com/nicolasseverino/xplia.git
cd xplia
pip install -e ".[full]"

---

⚡ Quick Start

30-Second Example

from xplia import create_explainer
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_iris

# Train a model
X, y = load_iris(return_X_y=True)
model = RandomForestClassifier()
model.fit(X, y)

# Create explainer (auto-detects best method for your model)
explainer = create_explainer(model, method='shap')

# Generate explanations
explanation = explainer.explain(X[:5])

# Access results
print(explanation.feature_importance)
print(explanation.quality_metrics)

Complete Production Example

from xplia import create_explainer, set_config
from xplia.compliance import GDPRCompliance, AIActCompliance
from xplia.explainers.trust import UncertaintyQuantifier, FairwashingDetector

# Configure XPLIA
set_config('verbosity', 'INFO')
set_config('n_jobs', -1)  # Use all CPU cores
set_config('cache_enabled', True)

# Create unified explainer (combines SHAP + LIME + Counterfactuals)
explainer = create_explainer(
    model,
    method='unified',
    methods=['shap', 'lime', 'counterfactual'],
    background_data=X_train
)

# Generate explanations
explanation = explainer.explain(X_test[:10])

# Check regulatory compliance
gdpr = GDPRCompliance(model, model_metadata={
    'name': 'Credit Scoring Model',
    'purpose': 'Loan approval',
    'legal_basis': 'legitimate_interest'
})
gdpr_report = gdpr.generate_dpia()
gdpr_report.export('gdpr_report.pdf')

ai_act = AIActCompliance(model, usage_intent='credit_scoring')
ai_act_report = ai_act.generate_compliance_report()

# Evaluate trust
uq = UncertaintyQuantifier(model, explainer)
uncertainty = uq.quantify(X_test)

detector = FairwashingDetector(model, explainer)
fairwashing = detector.detect(X_test, y_test)

# Generate comprehensive report
from xplia.visualizations import ChartGenerator
chart_gen = ChartGenerator()
chart_gen.create_dashboard(
    explanation,
    uncertainty=uncertainty,
    fairwashing=fairwashing,
    output='complete_report.html'
)

---

📚 Documentation

Comprehensive Guides

• Installation Guide - Detailed setup for all platforms
• Architecture - System design and patterns
• Plugin Development - Create custom explainers
• FAQ - Common questions and troubleshooting
• API Reference - Complete API documentation

Tutorials

• Explaining scikit-learn Models
• Explaining Deep Learning Models
• GDPR Compliance Workflow
• Production Deployment

---

💡 Examples

Real-World Use Cases

1. Loan Approval System (Complete Example)

python examples/loan_approval_system.py

Features:

• Model training and evaluation
• Multiple explanation methods
• GDPR and AI Act compliance
• Trust evaluation
• Audit trails
• Production-ready code

2. Healthcare Diagnosis Explainability

# See examples/healthcare_diagnosis.py

3. Fraud Detection with Counterfactuals

# See examples/fraud_detection.py

API Integration

REST API with FastAPI

from xplia.api import create_api_app

app = create_api_app(models={'my_model': model})

# Run with: uvicorn main:app --host 0.0.0.0 --port 8000

Endpoints:

• POST /explain - Generate explanations
• POST /compliance - Check compliance
• POST /trust/evaluate - Evaluate trust
• GET /health - Health check

MLflow Integration

from xplia.integrations.mlflow import XPLIAMLflowLogger

with XPLIAMLflowLogger(experiment_name="my_experiment") as logger:
    # Train model
    model.fit(X, y)

    # Log with explanations
    explainer = create_explainer(model)
    explanation = explainer.explain(X_test)
    logger.log_explanation(explanation)

Weights & Biases Integration

from xplia.integrations.wandb import XPLIAWandBContext

with XPLIAWandBContext(project="my-project") as logger:
    # Train and log
    model.fit(X, y)

    explainer = create_explainer(model)
    explanation = explainer.explain(X_test)
    logger.log_explanation(explanation)

---

🐳 Deployment

Docker

# Build image
docker build -t xplia:latest .

# Run API server
docker run -p 8000:8000 xplia:latest

# Or use docker-compose
docker-compose up

Kubernetes

# Deploy to Kubernetes
kubectl apply -f kubernetes/deployment.yaml

# Check status
kubectl get pods -l app=xplia

# Access API
kubectl port-forward svc/xplia-api-service 8000:80

Features:

• Horizontal Pod Autoscaling
• Health checks
• Persistent volumes
• Load balancing

---

🏗️ Architecture

High-Level Overview

User Interface (CLI, API, Notebooks)
            ↓
    Public API Layer
            ↓
     Core Framework
    (Factory, Registry, Config)
            ↓
┌───────────────────────────────────┐
│  Explanation Layer                │
│  (SHAP, LIME, Unified, etc.)     │
├───────────────────────────────────┤
│  Model Adapter Layer              │
│  (sklearn, TF, PyTorch, XGBoost) │
├───────────────────────────────────┤
│  Compliance & Trust Layer         │
│  (GDPR, AI Act, Uncertainty)     │
├───────────────────────────────────┤
│  Visualization Layer              │
│  (Charts, Reports, Dashboards)    │
└───────────────────────────────────┘

Key Design Patterns:

• Adapter Pattern - Unified interface for all ML frameworks
• Factory Pattern - Dynamic explainer creation
• Registry Pattern - Component discovery and versioning
• Strategy Pattern - Runtime algorithm selection

See ARCHITECTURE.md for complete details.

---

🧪 Testing

# Run all tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=xplia --cov-report=html

# Run specific test suite
pytest tests/explainers/ -v

# Run benchmarks
pytest tests/benchmarks/ -m benchmark

Test Coverage: 50%+ (6000+ lines of test code)

---

🤝 Contributing

We welcome contributions! Please see:

• Contributing Guide - How to contribute
• Code of Conduct - Community standards
• Development Setup - Get started

Quick Contribution Workflow

# Fork and clone
git clone https://github.com/YOUR-USERNAME/xplia.git

# Create branch
git checkout -b feature/amazing-feature

# Make changes, add tests, update docs

# Run tests
pytest tests/ -v

# Commit and push
git commit -m "feat: Add amazing feature"
git push origin feature/amazing-feature

# Open Pull Request on GitHub

---

📊 Comparison with Other Libraries

Feature	XPLIA	SHAP	LIME	Alibi	InterpretML
Methods	8+	1	1	5	2
GDPR Compliance	✅	❌	❌	❌	❌
Fairwashing Detection	✅	❌	❌	❌	❌
Multi-Audience	✅	❌	❌	❌	❌
REST API	✅	❌	❌	❌	❌
Uncertainty	✅	❌	❌	✅	❌
Production Ready	✅	⚠️	⚠️	✅	⚠️
Test Coverage	50%+	60%+	50%+	70%+	55%+

XPLIA Advantage: Only library with built-in compliance, fairwashing detection, and production deployment tools.

---

🗺️ Roadmap

v1.1.0 (Q1 2026)

• Interactive web dashboard (React)
• Additional compliance (SOC 2, ISO 27001)
• Automated model monitoring
• Advanced fairness metrics

v1.2.0 (Q2 2026)

• Distributed computing (Spark, Dask)
• Quantum ML explainability
• Federated learning support
• AutoML integration

v2.0.0 (Q3 2026)

• Causal inference integration
• Time series explainability
• Graph neural network support
• Multi-language support (R, Julia)

---

📜 License

XPLIA is licensed under the MIT License. See LICENSE for details.

---

📞 Support & Community

• Documentation: https://xplia.readthedocs.io
• GitHub Issues: https://github.com/nicolasseverino/xplia/issues
• Discussions: https://github.com/nicolasseverino/xplia/discussions
• Email: [e-mail]
• Twitter: @XPLIALib

---

🙏 Acknowledgments

XPLIA is built on the shoulders of giants:

• SHAP by Scott Lundberg
• LIME by Marco Tulio Ribeiro
• Alibi by Seldon
• InterpretML by Microsoft

Special thanks to all contributors and the open-source community.

---

📖 Citation

If you use XPLIA in your research, please cite:

@software{xplia2025,
  author = {Severino, Nicolas and contributors},
  title = {XPLIA: The Ultimate State-of-the-Art AI Explainability Library},
  url = {https://github.com/nicolasseverino/xplia},
  version = {1.0.0},
  year = {2025},
  doi = {10.5281/zenodo.XXXXXXX}
}

---

⭐ Star us on GitHub • 📦 Install Now • 📖 Read the Docs

Made with ❤️ by the XPLIA Team