Overview
VCM is a framework for generating synthetic data with controllable coverage guarantees. It ensures that generated samples adequately represent target subgroups and distributions in the real data.
Installation
pip install vcm-model
Quick Start
import vcm
model = vcm.VCM(coverage_target=0.95, min_group_size=50, diversity_weight=0.3)
model.fit(real_data, epochs=100)
synthetic_data = model.generate(n_samples=1000, coverage='guaranteed')
coverage = vcm.metrics.coverage(real_data, synthetic_data)
print(f"Actual coverage: {coverage:.2%}")
Coverage Modes
| Mode | Behavior |
|---|
guaranteed | Ensures minimum coverage across all subgroups |
balanced | Equal weight to coverage and fidelity |
fidelity | Optimizes for sample quality |
Dependencies
torch>=2.0numpy>=1.21scipy>=1.7scikit-learn>=1.0
Citation
@software{vcm,
title = {VCM: Variable Coverage Model},
author = {SSTDV Project},
year = {2024},
url = {https://github.com/SSTDV-Project/VCM}
}
← Back to Software References