faster_fishers

Fast implementation of Fisher's exact test in Rust for Python.
Benchmarks show that this version is about 30x faster than scipy's version when running on a large range of inputs and about 10x faster when running on 1 input:

---------------------------------------------------------------------------------------------- benchmark: 2 tests ---------------------------------------------------------------------------------------------
Name (time in ms)                        Min                   Max                  Mean             StdDev                Median                IQR            Outliers      OPS            Rounds  Iterations
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_benchmark_faster_fischer        59.8543 (1.0)         61.0522 (1.0)         60.5012 (1.0)       0.2816 (1.0)         60.5717 (1.0)       0.3114 (1.0)           3;1  16.5286 (1.0)          17           1
test_benchmark_scipy              1,859.7465 (31.07)    1,935.8237 (31.71)    1,885.3549 (31.16)    30.7295 (109.12)   1,871.9271 (30.90)    38.2479 (122.85)        1;0   0.5304 (0.03)          5           1
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Usage:

Rust:

use faster_fishers::{fishers_exact_with_odds_ratio, Alternative};
let table = [3, 5, 4, 50];
let alternative = Alternative::Less;
let (p_value, odds_ratio) = fishers_exact_with_odds_ratio(&table, alternative).unwrap();

Python:

>>> import numpy as np
>>> import fishers
>>> lefts, right, two_tails = fishers.exact(np.array([1, 3]), np.array([2, 5]), np.array([1, 4]), np.array([5, 50]))
>>> lefts
array([0.9166666666666647, 0.9963034765672586])
>>> rights
array([0.5833333333333326, 0.03970749246529451])
>>> two_tails
array([1.0, 0.03970749246529451])

Developing

Building with cargo

• Run cargo build in the main directory to build the project.

Publishing on pypi

docker run --rm -v $(pwd):/io ghcr.io/pyo3/maturin publish -u {USER} -p {PASSWORD}

Publishing on cargo

cargo publish

Using locally

• Install environment: poetry install
• Add environment to current shell poetry shell
• Install faster_fishers in current environment: maturin develop
• Check that it works: python -c "import faster_fishers; print(dir(faster_fishers))"

To try the library in a different environment:
1. maturin build --release
2. cd folder
3. pip install {wheel_path}.whl

Benchmarks

• Make sure to compile in release mode with maturin first: RUSTFLAGS='-C target-cpu=native' maturin develop --release *python: pytest --benchmark-warmup -m benchmark