datasentry 0.1.1

Data-centric ML inspection and auto-remediation toolkit

🚀 DataSentry

Proactive dataset validation for reliable machine learning systems.

DataSentry is a lightweight, production-focused Python library that detects structural dataset issues before model training begins.
It helps prevent silent performance degradation, misleading validation metrics, and costly deployment failures.

Because in machine learning --- bad data breaks good models.

---

🚩 Why DataSentry?

Modern ML pipelines frequently suffer from hidden dataset issues:

Problem Risk

---

⚖️ Class Imbalance Biased predictions 🏷 Label Noise Reduced generalization 🔓 Data Leakage Inflated validation accuracy 📉 Outliers Distorted feature space 🔄 Distribution Shift Poor real-world performance

DataSentry automatically identifies these risks early in your workflow.

---

🧠 Core Features

⚖️ Class Imbalance Detection

Evaluates label distribution and computes normalized imbalance metrics.

🏷 Label Noise Detection

Flags suspicious label inconsistencies affecting model learning.

🔓 Data Leakage Detection

Identifies features overly correlated with target variables.

📉 Outlier Detection

Detects anomalous samples that may distort training.

🔄 Distribution Shift Detection

Compares feature distributions to detect dataset drift.

---

📦 Installation

Install from PyPI:

pip install datasentry

Or install locally:

pip install .

---

⚡ Quick Start

import numpy as np
from datasentry import analyze

X = np.random.randn(100, 5)
y = np.array([0] * 90 + [1] * 10)

report = analyze(X=X, y=y)
print(report)

---

⚙ Advanced Usage

report = analyze(
    X=X,
    y=y,
    imbalance_threshold=3.0,
    outlier_threshold=0.1,
    leakage_threshold=0.9
)

---

🏗 Architecture

datasentry/
│
├── detectors/
│   ├── imbalance.py
│   ├── label_noise.py
│   ├── leakage.py
│   ├── outliers.py
│   └── shift.py
│
├── analyzer.py
├── config.py
├── report.py
├── utils.py
└── fixer.py

---

🧪 Running Tests

pytest

---

🗺 Roadmap

• CLI interface\
• Visualization dashboard\
• Advanced statistical leakage detection\
• Automated remediation suggestions\
• scikit-learn pipeline integration\
• Production drift monitoring

---

🤝 Contributing

Contributions are welcome. Please ensure:

• Clear documentation\
• Proper unit test coverage\
• Consistent coding standards\
• Descriptive commit messages

---

📄 License

MIT License

---

💡 Philosophy

Reliable machine learning begins with reliable data.

DataSentry focuses on structural dataset validation to reduce debugging effort, improve model robustness, and increase production stability.

v0.1.1

• Unified detector output schema
• Fixed imbalance threshold logic
• Refactored health score computation
• Fixed AutoFixer compatibility
• Updated test suite
• Improved package structure