valencity 0.1.3

A privacy & ML safety toolkit for production ML pipelines

🛡️ Valencity

The ML Safety Fortress
Privacy Engineering · Data Validation · Leakage Prevention

Why Valencity • Features • Install • Quick Start • Full API • CLI • Roadmap

"Stop shipping ML models that leak PII or data. Valencity catches it before you do."

---

🚀 Why Valencity?

Most ML teams discover data privacy problems in production — when it's already too late.
Valencity is the safety net that catches issues in your pipeline before they hit your users.

One line of code. Six layers of protection.

pip install valencity

Capability	Valencity 🛡️	Great Expectations	Pandera	Custom Scripts
PII Detection (50+ patterns)	✅	❌	❌	⚠️ Manual
Masking & Anonymization	✅	❌	❌	❌
K-Anonymity & Differential Privacy	✅	❌	❌	❌
ML Leakage Prevention	✅ Full Suite	❌	❌	❌
Data Drift Detection (4 methods)	✅	✅	❌	⚠️ Manual
Schema + Quality Checks	✅	✅	✅	⚠️ Manual
Synthetic Data Generation	✅	❌	❌	❌
GDPR / CCPA Compliance	✅	❌	❌	❌
HTML Reports	✅ Beautiful	✅	❌	❌
CLI Tool	✅	❌	❌	❌

---

✨ Features

🕵️ PII Detection & Anonymization

• Detect 50+ PII types — emails, phones, IBANs, passports, API keys, SSNs, IPs, and more
• Smart Masking: redact, hash (SHA-256), partial mask, or replace with realistic fake data
• NLP support for names and addresses via spaCy (optional)
• Async detector for non-blocking scans in production services
• Plugin-friendly: register your own custom PII patterns

✅ Data Validation & Quality

• Data Profiling: automatic statistical summary of any DataFrame
• Schema Validation: enforce strict data contracts — types, nullability, ranges
• Quality Checks: nulls, duplicates, outliers, cardinality, and more
• Fluent Expectations API: expressive, chainable validation rules
• Drift Detection: KS test, chi-squared, Jensen–Shannon divergence, and PSI

🚫 Leakage Prevention

• SafeCrossValidator: preprocessing only sees training fold — no more optimistic CV scores
• LeakageDetector: catches target leakage, train/test overlap, and temporal leakage
• SafeSplitters: time-series and group-aware train/test splits

🔐 Privacy Engineering

• Differential Privacy: add calibrated Laplace / Gaussian noise to statistics
• Compliance Checker: automated GDPR & CCPA violation detection with fix suggestions

🧬 Synthetic Data

• SyntheticGenerator: generate realistic fake datasets that mirror real data distributions

📊 Reporting

• Beautiful HTML reports for PII scans, quality audits, drift analysis, and compliance checks

---

📦 Installation

# Core (PII, validation, leakage, reports)
pip install valencity

# With NLP support (names, free-text addresses)
pip install valencity[nlp]
python -m spacy download en_core_web_sm

# Full developer install
pip install valencity[all]

Requirements: Python ≥ 3.9, pandas ≥ 1.5, scikit-learn ≥ 1.0

---

⚡ Quick Start

1 · Detect & Mask PII

import pandas as pd
from valencity.pii import PIIDetector, PIIMasker

df = pd.read_csv("users.csv")

# Scan
detector = PIIDetector()
report = detector.scan_dataframe(df)

if report.has_pii:
    print(f"⚠️  PII detected in columns: {report.pii_columns}")
    for col, findings in report.details.items():
        for f in findings:
            print(f"   [{f.pii_type}] → '{f.value}'")

# Mask
masker = PIIMasker(strategy="partial")   # [e-mail] → j***@example.com
safe_df = masker.mask_dataframe(df)

2 · Validate Schema & Quality

from valencity.validation import DataSchema, DataQualityChecker, DataProfiler

# Auto-infer schema from training data
schema = DataSchema.from_dataframe(train_df)
validation_result = schema.validate(production_df)

# Quality report
checker = DataQualityChecker()
quality = checker.check(df)
print(f"Null rate: {quality.null_rate:.1%} | Duplicates: {quality.duplicate_count}")

# Statistical profile
profiler = DataProfiler()
profile = profiler.profile(df)
print(profile.summary())

3 · Detect Data Drift

from valencity.validation import DriftDetector

detector = DriftDetector(method="ks_test")   # ks_test | chi2 | js | psi
detector.fit(reference_df=train_df)

drift = detector.detect(current_df=production_df)
if drift.has_drift:
    print(f"🚨 Drift detected in: {drift.drifted_columns}")

4 · Prevent Model Leakage

from valencity.leakage import SafeCrossValidator, LeakageDetector, SafeSplitter
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import StandardScaler

# Detect leakage before training
ld = LeakageDetector()
issues = ld.detect(X_train, X_test, y_train)
if issues:
    print(f"🚨 Leakage found: {issues}")

# Safe CV — preprocessing never sees test fold
cv = SafeCrossValidator(n_splits=5, preprocessor=StandardScaler())
scores = cv.cross_val_score(LogisticRegression(), X, y)
print(f"✅ Realistic accuracy: {scores.mean():.4f} ± {scores.std():.4f}")

# Safe time-series split
splitter = SafeSplitter(strategy="time_series")
X_train, X_test, y_train, y_test = splitter.split(X, y, time_col="date")

5 · Privacy Engineering

from valencity.privacy import DifferentialPrivacy, ComplianceChecker

# Add calibrated noise to protect aggregates
dp = DifferentialPrivacy(epsilon=1.0)
noisy_mean = dp.add_noise(df["salary"].mean(), sensitivity=1000)

# Check GDPR / CCPA compliance
checker = ComplianceChecker()
compliance = checker.check(df)
if not compliance.is_compliant:
    for violation in compliance.violations:
        print(f"❌ {violation.regulation}: {violation.description}")

6 · Generate Synthetic Data

from valencity.synthetic import SyntheticGenerator

gen = SyntheticGenerator()
synthetic_df = gen.generate(template_df=real_df, n_rows=1000)

7 · Generate HTML Reports

from valencity.reports import HTMLGenerator
from valencity.pii import PIIDetector

detector = PIIDetector()
pii_report = detector.scan_dataframe(df)

gen = HTMLGenerator()
gen.generate_pii_report(pii_report, output_path="pii_report.html")
print("📊 Report saved → pii_report.html")

---

🖥️ Full API

import valencity
valencity.show_api()

Module	Class / Function	Purpose
valencity.pii	PIIDetector	Scan DataFrames & text for PII
valencity.pii	PIIMasker	Mask / anonymize PII
valencity.pii	PIIType	Enum of 50+ supported PII types
valencity.pii	PIIPatterns	Pattern registry (plugin-friendly)
valencity.pii	AsyncPIIDetector	Async, non-blocking PII scanning
valencity.validation	DataSchema	Schema validation & contracts
valencity.validation	DataQualityChecker	Null / duplicate / outlier checks
valencity.validation	DataProfiler	Statistical profiling
valencity.validation	DriftDetector	KS, chi2, JS, PSI drift detection
valencity.validation	expect()	Fluent expectations API
valencity.leakage	SafeCrossValidator	Leakage-safe cross-validation
valencity.leakage	SafeSplitter	Time-series & group-aware splits
valencity.leakage	LeakageDetector	Target, overlap & temporal leakage
valencity.privacy	DifferentialPrivacy	Laplace / Gaussian DP noise
valencity.privacy	ComplianceChecker	GDPR / CCPA violation detection
valencity.synthetic	SyntheticGenerator	Realistic synthetic data
valencity.reports	HTMLGenerator	Beautiful HTML audit reports

---

🖥️ CLI

# Show package info
valencity info

# Scan a CSV for PII
valencity pii scan users.csv

# Mask PII in a CSV
valencity pii mask users.csv --output safe_users.csv

# Generate a data profile
valencity profile dataset.csv

# Check GDPR/CCPA compliance
valencity compliance dataset.csv

# Run quality checks
valencity validate quality dataset.csv

---

🗺️ Roadmap

• PII Detection — 50+ patterns, masking, async support
• Data Validation — schema, quality, drift, profiling
• Leakage Prevention — CV, splitters, detectors
• Privacy Engineering — differential privacy, compliance
• Synthetic Data Generation
• CLI Tool
• MLflow Integration — auto-log safety checks to experiments
• Airflow Operator — drop-in pipeline safety checks
• dbt Integration — run Valencity checks in dbt tests
• Dashboard UI — real-time data safety monitoring

---

🤝 Contributing

Contributions are welcome! Check the open issues.

git clone https://github.com/ihakawati/valencity
cd valencity
pip install -e ".[dev]"
pytest

1. Fork the repo
2. Create a feature branch: git checkout -b feature/my-feature
3. Commit: git commit -m "feat: add my feature"
4. Push: git push origin feature/my-feature
5. Open a Pull Request

---

Built with ❤️ for the ML Community

PyPI · GitHub · Report a Bug