autoparquet 0.1.5

Automatically generate optimized Parquet/Arrow schemas for better compression.

AutoParquet

AutoParquet is a Python package that wraps Parquet/Arrow to automatically generate optimized schemas for your data. It focuses on better compression through automatic bit-packing, int-packing, and dictionary encoding, while providing a convenient "header" system for storing custom metadata.

Quick Start

Installation

git clone https://github.com/edawson/autoparquet
cd autoparquet
pip install -e .

Or with dev dependencies:

pip install -e ".[dev,polars]"

Basic Example

import pandas as pd
import autoparquet

df = pd.DataFrame({
    "id": [1, 2, 3, 4, 5],
    "category": ["A", "B", "A", "B", "A"],
    "value": [1.1, 2.2, 3.3, 4.4, 5.5]
})

# Write with automatic schema optimization and a custom header
autoparquet.write_parquet(
    df,
    "data.parquet",
    header={"version": "1.0", "author": "Eric T. Dawson"}
)

# Read back the data and the header
df_read, header = autoparquet.read_parquet("data.parquet")
print(header)  # {'version': '1.0', 'author': 'Eric T. Dawson'}

Command Line

# Convert CSV to optimized Parquet (default: zstd compression)
autoparquet csv_to_parquet data.csv

# Convert CSV to Feather with custom options
autoparquet csv_to_feather data.csv -o out.feather -c snappy

# Tab-separated input with reduced float precision
autoparquet csv_to_parquet data.tsv -d $'\t' -f float32

Features

• Automatic Schema Inference: Downcasts integers to the smallest type that fits and optionally reduces float precision.
• Optimized Compression: Dictionary-encodes low-cardinality strings with the smallest index type; converts uniform-length strings to FixedSizeBinary.
• Custom Headers: Easily add and retrieve custom metadata (versioning, key-value pairs) in Parquet files.
• Multi-Framework Support: Works with Pandas, Polars, and cuDF.
• CLI: Convert CSV files to optimized Parquet or Feather from the command line.

Documentation

• Usage Guide - Detailed examples, API reference, and advanced features
• Contributing - Development setup and contribution guidelines

Development

# Run tests
pytest

# Lint and format
ruff check .
ruff format .

# Type checking
mypy src/autoparquet tests

# Or use make
make test
make lint
make check

Requirements

• Python 3.9+
• PyArrow 14.0.0+
• Pandas 2.0.0+ (optional)
• Polars 0.20.0+ (optional)

License

MIT License

Citation

@software{autoparquet2026,
  author = {Dawson, Eric T.},
  title = {AutoParquet: Automatic Schema Optimization for Parquet Files},
  year = {2026},
  url = {https://github.com/edawson/autoparquet}
}

---

This repository was generated using Gemini 3 Flash, based on a specification written by the author. The code was reviewed by the author for correctness and tested locally.