Redbear

A high-performance Python extension for fast element-wise dictionary operations. Redbear provides a minimal, purpose-built RedDict class that delivers significant performance improvements over traditional data science libraries for specific use cases involving numeric dictionaries.

Purpose

Redbear is inspired by blackbear and builds on the concepts introduced there. By moving the implementation to Rust while keeping things as simple as possible, Redbear achieves far better performance across the board.

The library is designed for scenarios where:

• You work with numeric dictionaries (string keys, float values)
• You need fast element-wise operations (add, subtract, multiply)
• Performance matters more than generality

After heavy optimizations using PyO3 and Arc-based internal sharing, Redbear achieves excellent performance while hiding all the complexity from the library users.

Installation

pip install redbear

Usage

import redbear as rb

# Create a RedDict from a regular Python dict.
data = {"a": 1.0, "b": 2.0, "c": 3.0}
rd = rb.RedDict(data)

# Scalar operations (creates a new RedDict).
rd_plus_5 = rd.add_scalar(5.0)  # {"a": 6.0, "b": 7.0, "c": 8.0}
rd_minus_2 = rd.subtract_scalar(2.0)  # {"a": -1.0, "b": 0.0, "c": 1.0}

# Element-wise operations between two RedDicts.
other = rb.RedDict({"a": 10.0, "b": 20.0, "c": 30.0})
result = rd.add(other)  # {"a": 11.0, "b": 22.0, "c": 33.0}
result = rd.subtract(other)  # {"a": -9.0, "b": -18.0, "c": -27.0}
result = rd.multiply(other)  # {"a": 10.0, "b": 40.0, "c": 90.0}

# Perform item lookup. Note that it's up to 2x slower than Python's dict item lookup.
data["a"] # 1.0

# Get the underlying dict back.
plain_dict = rd.to_dict  # {"a": 1.0, "b": 2.0, "c": 3.0}

Benchmarks

Run benchmarks with:

pip install -e ".[dev]"
python -m benchmarks.redbear_way
python -m benchmarks.numpy_way
python -m benchmarks.blackbear_way
python -m benchmarks.polars_way
python -m benchmarks.pandas_way

Best Case (keys identical)

Library	Operations	Collection Size	Time (seconds)
Redbear	100000 x 5	10	0.056
Redbear	1000000 x 5	10	0.583
Redbear	10000 x 5	1000	0.015
Redbear	100000 x 5	1000	0.145
Numpy	100000 x 5	10	0.285
Numpy	1000000 x 5	10	2.436
Numpy	10000 x 5	1000	0.039
Numpy	100000 x 5	1000	0.371
Blackbear	100000 x 5	10	0.371
Blackbear	1000000 x 5	10	3.755
Blackbear	10000 x 5	1000	3.643
Blackbear	100000 x 5	1000	36.307
Polars	100000 x 5	10	1.072
Polars	1000000 x 5	10	11.147
Polars	10000 x 5	1000	0.132
Polars	100000 x 5	1000	1.219
Pandas	100000 x 5	10	13.466
Pandas	1000000 x 5	10	130.837
Pandas	10000 x 5	1000	1.339
Pandas	100000 x 5	1000	13.432

Worst Case (different key order)

Library	Operations	Collection Size	Time (seconds)
Redbear	100000 x 5	10	0.561
Redbear	1000000 x 5	10	5.432
Redbear	10000 x 5	1000	0.283
Redbear	100000 x 5	1000	2.842

Development

Start developing with maturin:

maturin develop