Python packages

AGENT supports a comprehensive collection of privacy-preserving Python libraries, each wrapped with the op_ prefix to ensure differential privacy guarantees. These libraries enable data scientists to perform machine learning, statistical analysis, and data manipulation tasks while maintaining strict privacy protections through controlled noise injection and privacy budget management.

Pandas

Comprehensive data manipulation guide covering PrivateDataFrame and PrivateSeries with tutorials on importing, managing, operations, and statistical methods.

Diffprivlib

Differential privacy machine learning library providing scikit-learn compatible algorithms for regression, classification, and preprocessing with epsilon budget management.

OpenDP

Modular differential privacy framework for building custom pipelines with transformations, measurements, and automatic privacy budget tracking.

SmartNoise SQL

SQL interface for differential privacy queries on PrivateDataFrames using the SmartNoise framework with Privacy and Reader objects.

SmartNoise Synth

Synthetic data generation library creating privacy-protected tabular datasets while preserving statistical properties and relationships.

Opacus

PyTorch differential privacy library with PrivateDPDataLoader integration for training neural networks using Privacy Engine and DP-SGD optimization.

TensorFlow Privacy

TensorFlow/Keras integration for differential privacy using PrivateKerasModel wrapper and DP-SGD training with support for model import/export.

Record Linkage

Privacy-aware record linkage toolkit for identifying and deduplicating records across datasets while protecting individual privacy.

SPlink

Probabilistic record linkage library for accurate entity resolution and data integration with customizable matching rules and privacy preservation.