sdg-core-lib 0.1.9.dev32

Advanced synthetic data generation library

GENESIS Core Lib

What is GENESIS Core Lib?

GENESIS Core Lib is an open-source, advanced synthetic data generation library for Python 3.12+ that provides state-of-the-art machine learning models including VAEs (TabularVAE, TimeSeriesVAE) and CTGAN for generating high-quality tabular and time series data. The library features adaptive on-the-fly training with automatic model adaptation to data characteristics, model persistence for reusing trained models, behavior control through custom mathematical functions, and integrated quality evaluation metrics. Designed with privacy preservation in mind and optimized for both CPU and GPU processing, it offers a comprehensive Job API for data augmentation, privacy preservation, and ML model testing across various data types and use cases.

Why use GENESIS core Lib?

GENESIS Core Lib is the ideal solution for synthetic data generation because it specializes in industrial sensor data and time series applications, offering state-of-the-art models like TimeSeriesVAE that preserve temporal patterns and statistical properties essential for real-world scenarios. It provides immediate access to high-quality synthetic data without the costs and delays of physical sensor deployments, enabling rapid prototyping, algorithm development, and comprehensive testing while maintaining data privacy through synthetic data sharing. The library ensures data fidelity with advanced evaluation metrics including Dynamic Time Warping and correlation preservation, making it perfect for manufacturing IoT, environmental monitoring, energy utilities, and medical device applications where realistic temporal data is critical.

Nyquist plots created merging two time-series of real and imaginary impedances. Can you tell the difference?

✨ Key Features

• Generative AI Architectures: Advanced VAEs (TabularVAE, TimeSeriesVAE) and CTGAN for Tabular and Time series data
• Adaptive Training: On-the-fly model training with automatic adaptation to your data characteristics
• Model Persistence: Save and reuse trained generative models for consistent data generation
• Behavior Control: Manipulate generation patterns with custom mathematical functions
• Integrated Evaluation: Built-in quality assessment metrics for comprehensive data evaluation
• High Performance: Optimized for both CPU and GPU processing

🛠️ Quick Start

Quick Install

pip install sdg-core-lib

🚀 Try it

from sdg_core_lib import Job

# Text-based JSON configuration (no file needed)
config = {
    "n_rows": 1000,
    "model": {
        "algorithm_name": "sdg_core_lib.data_generator.models.VAEs.implementation.TabularVAE.TabularVAE",
        "model_name": "customer_synthetic_model"
    },
    "dataset": {
        "dataset_type": "table",
        "data": [
            {
                "column_data": [13.71, 13.4, 13.27, 13.17, 14.13, 13.88, 13.24, 13.73],
                "column_name": "alcohol",
                "column_type": "continuous",
                "column_datatype": "float64"
            },
            {
                "column_data": [5.65, 3.91, 4.28, 2.59, 4.1, 3.9, 3.8, 4.2],
                "column_name": "malic_acid",
                "column_type": "continuous",
                "column_datatype": "float64"
            },
            {
                "column_data": [1.28, 1.05, 1.02, 1.03, 1.71, 1.23, 1.07, 1.5],
                "column_name": "ash",
                "column_type": "continuous",
                "column_datatype": "float64"
            }
        ]
    },
    "save_filepath": "./models"
}

# Create and run a synthetic data generation job
job = Job(
    n_rows=config["n_rows"],
    model_info=config["model"],
    dataset=config["dataset"],
    save_filepath=config.get("save_filepath", "./models")
)

# Generate synthetic data
results, metrics, model, schema = job.train()
print(f"Generated {len(results)} synthetic rows")
print(f"Quality metrics: {metrics}")

📖 See Quick Start Guide for detailed examples

📚 Documentation

📖 User Documentation

Complete guide for users including:

• Core concepts and the Job API
• Data Types and Datasets
• Fantastic Models and how to use them
• How to handle raw data with Processors
• How to control generation of synthetic data with Functions
• Evaluate your work with Evaluators

🔧 Developer Documentation

Technical documentation for developers:

• Architecture overview and design patterns
• Extension points and customization
• Development setup and testing
• Code organization and standards

⚡ Quick Start Guide

Get started immediately with:

• Installation instructions
• Basic examples and tutorials
• Common use cases
• Troubleshooting tips

📋 Standalone Installation

Info about installation:

• Requirements
• Installation instructions
• Environment Variables
• GPU Support

Roadmap

A detailed Roadmap can be found following the Roadmap Link

TL;DR

Pre-1.0: Enhanced model quality, flexible architectures, advanced evaluation metrics, and expansion to images/text generation with auto hyperparameter tuning.

Post-1.0: Production-ready stability, mixed data types, intelligent automation, cloud integration, and advanced AI capabilities including federated learning and quantum computing exploration.

🤝 Contributing

We welcome contributions! Please see our Contributing Guidelines for details.

📄 License

This project is licensed under the GNU Affero General Public License v3.0 - see the LICENSE file for details.

🙏 Acknowledgments

• Built with TensorFlow and Keras for deep learning models
• Statistical evaluation using scipy and numpy
• Inspired by state-of-the-art synthetic data generation research

📞 Support

• 📖 Documentation
• 🐛 Issues
• 💬 Discussions

---

GENESIS Core Lib - Generating Tomorrow's Data, Today 🚀