SFDAO - Synthetic Finance Data Auditor & Optimizer

Financial Compliance & Synthetic Data Quality Assurance Platform

日本語版 (Japanese)

Overview

SFDAO is an integrated tool for synthetic data generation, constraint application, and auditing specifically designed for the financial industry. Covering Phases 1 to 3, it handles not only auditing but also generation, guardrail checking, scenario injection, and ML Utility evaluation.

Key Features

• Statistical Quality Evaluation: Distribution comparison using KS test and Jensen-Shannon Divergence.
• Finance-Specific Evaluation: Fat Tail detection, Volatility Clustering verification.
• Privacy Evaluation: Re-identification risk, Distance to Closest Record (DCR).
• Automated Type Detection: Automatic classification of Numeric, Categorical, Datetime, and PII (Personally Identifiable Information).
• Generation Workflow: Batch execution of synthetic data generation and auditing via generate/run.
• Constraints & Scenarios: Guardrail rule application, scenario injection (scale/shift/clip/outlier, etc.).
• ML Utility Evaluation: Model performance assessment using TSTR (AUC/F1) (optional).
• Report Generation: Detailed reports in HTML/PDF formats.

Installation

Quick Install (PyPI)

# Install via pip
pip install sfdao

# Or use pipx for isolated installation (recommended)
pipx install sfdao

# With optional deep learning support (CTGAN)
pip install sfdao[deep]

Prerequisites

• Python 3.10 - 3.12
• macOS: WeasyPrint dependencies for PDF generation

  brew install cairo pango gdk-pixbuf libffi
  

Development Setup

# Clone the repository
git clone https://github.com/takurot/sfdao.git
cd sfdao

# Install Poetry (if not already installed)
curl -sSL https://install.python-poetry.org | python3 -

# Install dependencies
poetry install

# Enable virtual environment
poetry shell

Quick Start

# Run a basic audit
sfdao audit --real data/real.csv --synthetic data/synthetic.csv --output report.html

# Output format is automatically detected by extension (.txt/.html/.pdf)
sfdao audit --real data/real.csv --synthetic data/synthetic.csv --output report.txt
sfdao audit --real data/real.csv --synthetic data/synthetic.csv --output report.pdf

# Generate simple synthetic data for testing
poetry run python -m sfdao.scripts.generate_test_synthetic_data \
  example/data/creditcard_real_sample.csv \
  example/output/creditcard_synthetic.csv \
  --n-samples 500 \
  --random-state 42

# Audit the generated synthetic data
poetry run sfdao audit \
  --real example/data/creditcard_real_sample.csv \
  --synthetic example/output/creditcard_synthetic.csv \
  --output example/output/report.html

# Phase 2: Batch execution of Generation -> Guardrails -> Audit
poetry run sfdao run --config example/config/phase2.yaml --outdir example/output

Development

TDD (Test-Driven Development)

This project is developed using TDD. Follow this cycle when adding new features:

1. Red: Write a failing test.
2. Green: Write the minimum code to pass the test.
3. Refactor: Clean up and optimize the code.

Testing

# Run all tests
pytest

# Run with coverage report
pytest --cov=sfdao --cov-report=html

# Run specific test file
pytest tests/unit/ingestion/test_loader.py

Code Quality

# Check formatting
black --check .

# Apply formatting
black .

# Lint check
flake8 .

# Type check
mypy sfdao

# Security check
bandit -r sfdao

Project Structure

sfdao/
├── sfdao/                  # Main package
│   ├── ingestion/          # Data ingestion and type detection
│   ├── config/             # Configuration schema/loader
│   ├── generator/          # Synthetic data generation
│   ├── guard/              # Rule-based constraint checking
│   ├── scenario/           # Scenario injection
│   ├── evaluator/          # Metric calculation
│   ├── reporter/           # Report generation
│   └── cli/                # CLI interface
├── tests/                  # Test code
│   ├── unit/               # Unit tests
│   ├── integration/        # Integration tests
│   └── e2e/                # End-to-End tests
├── docs/                   # Documentation
└── prompt/                 # Specifications

Documentation

• Implementation Plan
• Product Specifications
• Example
• Usage Guide
• Architecture
• Python API
• Metrics Guide

Roadmap

Phase 1: "The Auditor" (MVP)

• Project structure and CI/CD setup
• Basic Data Ingestion features
• Auto-Type Detection
• Finance Domain definitions
• Basic Evaluator (statistical tests)
• Financial Stylized Facts evaluation
• Privacy evaluation
• Evaluation scoring integration
• CLI interface
• Report generation feature
• Integration testing and documentation

Phase 2: "The Generator & Logic"

• Config schema/loader and CLI integration (generate/run)
• Baseline Generator (statistical sampling)
• Constraint & Logic Guard (rule detection/exclusion/correction)
• Scenario Injection (scale/shift/clip/outlier, etc.)
• E2E workflow (generate -> guard -> audit)
• Benchmark and Privacy sampling

Phase 3: "The Professional"

• CI/CD optimization and Release workflow
• Advanced Generator (CTGAN, optional)
• ML Utility evaluation (TSTR: AUC/F1)
• PyPI metadata/CHANGELOG/README maintenance

Future Ideas

• Rule Learning Engine (Reinforcement Learning based)
• Auto-Tuning Mode (Autonomous quality improvement)

Contributing

Contributions are welcome! Please follow these steps:

1. Fork this repository.
2. Create a feature branch (git checkout -b feature/amazing-feature).
3. Write tests before implementing (TDD).
4. Commit your changes (git commit -m 'Add amazing feature').
5. Push to the branch (git push origin feature/amazing-feature).
6. Create a Pull Request.

License

MIT License - see the LICENSE file for details.

Contact

For questions or suggestions regarding the project, please create an Issue.

Acknowledgments

• SDV (Synthetic Data Vault)
• CTGAN
• Kaggle Credit Card Fraud Detection Dataset