Synthetic Data SDK

Python client library for the Synthetic Data REST API. Generate high-fidelity synthetic data remotely without running heavy ML models locally.

Requirements

• Running Synthetic Data Server (for API endpoint)

Installation

# Using pip
pip install -e .

# Using uv, by adding dependency
uv sync

Features

• Remote synthesis: No local GPU or heavy dependencies required
• Multiple models: VineCopula, TabularGAN, TabDiff, SMOTE
• Multi-table support: Preserve foreign key relationships
• Privacy evaluation: Built-in privacy attack simulations (MIA, SARP, linkage attacks)
• Causal fidelity evaluation: Validate treatment effects, decision consistency, fairness preservation
• Pandas integration: Works seamlessly with DataFrames
• Quality evaluation: Built-in metrics for fidelity assessment (KS test, correlation, TSTR/TRTR)
• Model inspection: Access model metadata and configuration via summary()
• Unified API: Consistent interface across all synthesizers

Quick Start

Synthetic Data Generation

from synthetic_data_sdk import RemoteVineCopula
import pandas as pd

# Load your data
data = pd.read_csv("customers.csv")

# Initialize remote synthesizer
synth = RemoteVineCopula(
    endpoint="http://localhost:8000",
    model_version="customer-synth-v1"
)

# Fit model (runs on remote server)
synth.fit(data)

# Generate synthetic data
synthetic_data = synth.transform(n=1000)

# Inspect model configuration
summary = synth.summary()
print(f"Fitted: {summary['fitted']}")
print(f"Continuous cols: {summary['n_continuous']}")

# Evaluate synthetic data quality
metrics = synth.evaluate(
    real_data=data,
    synthetic_data=synthetic_data,
    categorical_cols=['city'],
    target_col='churn',
    task_type='classification'
)
print(f"Correlation Error: {metrics['correlation_error']:.3f}")
print(f"TSTR Score: {metrics['tstr_score']:.3f}")

Privacy Risk Evaluation

from synthetic_data_sdk import PrivacyEvaluator
import pandas as pd

# Load datasets
train_data = pd.read_csv("train.csv")
test_data = pd.read_csv("test.csv")
synthetic_data = pd.read_csv("synthetic.csv")

# Initialize privacy evaluator
privacy = PrivacyEvaluator(endpoint="http://localhost:8000")

# Run privacy attack simulations
results = privacy.evaluate(
    train_real_data=train_data,
    test_real_data=test_data,
    synthetic_data=synthetic_data,
    sensitive_columns=['ssn', 'salary', 'diagnosis']
)

# Check results
print(f"Overall Privacy Risk: {results['overall_risk']}")
print(f"Successful Attacks: {results['summary']['successful_attacks']}/{results['summary']['total_attacks']}")

# Review individual attacks
for attack in results['attacks']:
    print(f"{attack['attack_type']}: {attack['risk_level']} risk")

Causal Fidelity Evaluation

from synthetic_data_sdk import CausalEvaluator
import pandas as pd

# Load datasets
real_data = pd.read_csv("real.csv")
synthetic_data = pd.read_csv("synthetic.csv")

# Initialize causal evaluator
causal = CausalEvaluator(endpoint="http://localhost:8000")

# Evaluate treatment effect preservation
results = causal.evaluate(
    real_data=real_data,
    synthetic_data=synthetic_data,
    treatment_col='treatment_assigned',
    outcome_col='outcome_value',
    covariates=['age', 'income'],
    evaluation_type='treatment_effect'
)

# Check results
print(f"Overall Preserved: {results['overall_preserved']}")
print(f"ATE Real: {results['evaluations'][0]['metrics']['ate_real']:.3f}")
print(f"ATE Synthetic: {results['evaluations'][0]['metrics']['ate_synth']:.3f}")
print(f"Relative Error: {results['evaluations'][0]['metrics']['ate_relative_error']:.1%}")

Available Synthesizers

Documentation

API reference is available via docstrings, also refer to Online Documentation

Support

• Issues: Report bugs and request features via issue tracker
• Email: info@protegrity.com

License

MIT License