data-prep-engine 0.1.0

A unified data ingestion and sanitization engine for ML workflows.

data-prep-engine

A lightweight, modular, production-friendly Python library for data ingestion, diagnostics, sanitization, visualization, and end-to-end ML data preparation.

Built to provide a single, unified, reproducible pipeline that works across CSV, JSON, Parquet, images, and more — without depending on massive profiling libraries.

---

🌟 Key Features

🔌 1. Ingestion Engine (Loader)

Load CSV, JSON, Parquet, Images, and more into a unified StandardTable.

🩺 2. Diagnostics Engine (DataDoctor)

Column-level summaries, warnings, null counts, duplicates, outliers, cardinality & constant-column detection.

🧼 3. Sanitization Engine (The Surgeon)

• Missing value imputation
• Duplicate removal
• IQR-based outlier capping
• Fully extensible sanitization steps

🎨 4. Visualization Engine (The Artist)

Smart single-flag plots (numeric histograms, categorical counts) guided by the diagnostics report.

🚀 5. AutoPrep — End-to-End Unified Pipeline

One line to prepare any dataset:

from data_prep_engine import AutoPrep

prep = AutoPrep.default()
result = prep.run_from_uri("data.csv")

result.cleaned_table.to_pandas().head()

⚙️ Installation

```bash
pip install data-prep-engine

(Coming soon to PyPI — currently install locally using:)

### pip install -e .

🏁 Quickstart

from data_prep_engine import AutoPrep
prep = AutoPrep.default()
# Load, diagnose, clean, visualize — all in one step
result = prep.run_from_uri("data.csv")
print(result.sanitization_logs)
result.cleaned_table.to_pandas().head()

📥 Ingestion Examples

from data_prep_engine.ingestion import Loader
loader = Loader()
table_csv    = loader.load("data.csv")
table_json   = loader.load("data.json")
table_parquet = loader.load("data.parquet")
table_image   = loader.load("image.jpg")   # stored as array metadata

All ingestion results are returned as a StandardTable, guaranteeing uniform structure.

🩺 Diagnostics Examples

from data_prep_engine.diagnostics import DataDoctor
doctor = DataDoctor()
report = doctor.diagnose(table)
print(report.summary_table())
print(report.warnings)

Common warnings include: • High missing values • High-cardinality categorical columns • Outlier-heavy numeric columns • Constant features Duplicate rows

🧼 Sanitization Examples

from data_prep_engine.sanitization.pipeline import SanitizationPipeline
from data_prep_engine.sanitization.steps import (
    MissingValueHandler,
    DuplicateHandler,
    OutlierHandler,
)
pipeline = SanitizationPipeline([
    MissingValueHandler(),
    DuplicateHandler(),
    OutlierHandler(),
])
result = pipeline.run(table)
clean_table = result.cleaned_table
print(result.logs)

🎨 Visualization Examples

from data_prep_engine.visualization import Artist
fig = Artist.plot(clean_table, doctor.diagnose(clean_table))
fig.show()
Or save as PNG:

Artist.to_png(fig, "preview.png")

🚀 Full AutoPrep Pipeline

from data_prep_engine import AutoPrep
prep = AutoPrep.default()
result = prep.run_from_uri("data.csv")
print("Warnings before:", result.diagnostics_before.warnings)
print("Warnings after:", result.diagnostics_after.warnings)
fig = prep.plot(result)
fig.show()

###📐 Project Architecture

data_prep_engine/
│
├── ingestion/         # Loaders + format adapters
├── diagnostics/       # DataDoctor + reports
├── sanitization/      # Sanitization steps + pipeline
├── visualization/     # Artist plotting engine
├── core/              # StandardTable + utilities
└── autoprep.py        # Full unified pipeline

Each block is independent, testable, and extendable.

🧪 Running Tests

pytest -q

The test suite covers: • Ingestion adapters • Diagnostics summaries • Sanitization steps • Visualization engine • AutoPrep unified pipeline

🤝 Contributing

1. Fork the repo
2. Create a feature branch
3. Add tests for your change
4. Submit a PR

All PRs must pass GitHub Actions (unit tests, linting, and security scanners).

📜 License

MIT License © 2025