sanitify 0.1.0

Intelligent Data Quality Analysis and ML-Assisted Data Cleaning

SanitiPy

Intelligent data quality analysis and ML-assisted data cleaning for production Python workflows.

SanitiPy is a production-grade Python library built to systematically analyze, validate, score, and improve structured datasets before they enter analytics or machine learning pipelines.

It provides structured profiling, rule-based quality validation, explainable quality scoring, deterministic cleaning utilities, and ML-assisted fix suggestions — all through a single public entry point:

from sanitipy import DataCleaner

---

🚀 Why SanitiPy?

Modern data systems fail more often due to poor data quality than model limitations.

SanitiPy helps teams:

• Detect structural issues early
• Quantify dataset health
• Enforce quality standards
• Apply deterministic cleaning safely
• Receive ML-assisted improvement suggestions
• Maintain transparency and reproducibility

Designed for:

• Data Engineers
• ML Engineers
• Analytics Teams
• Data Platform Teams
• Startups building internal data tooling

---

🧠 Design Philosophy

SanitiPy follows strict engineering principles:

• Single public API — DataCleaner
• No hidden mutations — data is never altered silently
• ML never auto-applies fixes — human-in-the-loop by design
• Deterministic-first approach — rules before models
• Explainable scoring logic
• Modular architecture
• Production-ready src layout
• Test-covered implementation

---

📦 Core Features (v1.0)

• Structured dataset profiling (schema-aware)
• Scalable sampling for large datasets
• Column-level metadata extraction
• Rule-based quality validation engine
• Explainable weighted quality scoring
• Deterministic cleaning operations
• ML-assisted fix suggestions (confidence-based)
• Structured JSON report export
• Clean, modular architecture

---

⚙️ Installation

Development Install

pip install -e ".[dev]"

Future PyPI Install

pip install sanitipy

---

🔍 Quick Example

import pandas as pd
from sanitipy import DataCleaner

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

dc = DataCleaner(df)

# 1. Profile dataset
profile = dc.profile()

# 2. Detect quality issues
issues = dc.check_quality()

# 3. Compute explainable quality score
score = dc.quality_score()

print(score)

---

📊 Example Output

Quality Issues

[
  {
    "column": "age",
    "rule": "high_missing",
    "severity": "medium",
    "metric": 0.42,
    "threshold": 0.3
  }
]

Quality Score

{
  "score": 72,
  "max_score": 100,
  "penalties": [
    {"rule": "high_missing", "deduction": 20}
  ]
}

Fully transparent. Fully explainable.

---

🏗 Architecture Overview

SanitiPy uses a modular, production-oriented structure:

sanitipy/
    datacleaner.py   # Public API
    core/            # Profiling, rules, scoring
    cleaning/        # Deterministic cleaning
    ai/              # ML suggestions
    report/          # Structured exports
    utils/           # Validation & helpers

Architectural Principles

• src/ layout for clean packaging
• Clear separation of concerns
• Rule engine abstraction
• Configurable scoring engine
• No visualization inside core
• No notebook dependencies
• Stable output schemas

---

🧮 Quality Engine

SanitiPy includes a rule-based validation system:

Built-in rules (v1):

• High missing rate detection
• Constant column detection
• High cardinality detection
• High duplicate rate detection

The engine is extensible and designed for future plugin support.

---

🔐 ML-Assisted Suggestions

SanitiPy supports ML-driven fix recommendations.

Key guarantees:

• Suggestions are confidence-scored
• Fixes are never auto-applied
• Users must explicitly approve changes
• Deterministic cleaning remains primary

---

🛣 Roadmap

• Structured profiling engine
• Rule-based validation engine
• Weighted scoring engine
• Deterministic cleaning utilities
• ML suggestion engine
• Report exporters (JSON/YAML)
• Streamlit demo application
• PyPI release

---

🧪 Development

Run tests:

pytest

Run with coverage:

pytest --cov=sanitipy --cov-report=term-missing

SanitiPy follows:

• Test-driven development
• Modular design
• Public API stability
• Semantic commit conventions

---

🤝 Contributing

Contributions are welcome.

Before submitting a PR:

• Ensure tests pass
• Maintain coverage
• Follow existing architecture patterns
• Avoid breaking public API
• Keep changes modular

---

📜 License

MIT License

---

🔮 Vision

SanitiPy aims to become a lightweight but powerful data quality foundation layer for modern Python data stacks — sitting between raw ingestion and analytics/ML pipelines.

Transparent. Deterministic. Extensible. Production-ready. "# trigger CI"