fisher-rxc

Fisher RxC

Fast multithreaded implementation of calculating Fisher's exact test for any RxC size table. Written in Rust using Maturin.

Installation

pip install fisher-rxc

import fisher

Usage

fisher.exact(table, workspace=None)

Calculate Fisher's exact test for 2D list according to Mehta & Patel's Network Algorithm. If workspace size is not provided, it will be "guessed" dynamically.

Workspace size of 2e8 takes at most 800MB of RAM.

fisher.sim(table, iterations)

Calculate Fisher's exact test for 2D list by multithreaded Monte Carlo simulation. A modern CPU can quickly do 10^7 iterations and get accurate results.

fisher.recursive(table)

[!WARNING] This is experimental and may not work on every CPU platform and generation. Any feedback is appreciated

Calculate Fisher's exact test by a multithreaded SIMD recursive algorithm. Despite extensive optimization efforts, this is still generally much slower than the fisher.exact function. Only use for small tables with low numbers.

Return values

0 <= x <= 1: p-value

x < 0: error code number, message printed to stdout

Performance

See benchmark.py

AMD Ryzen 5600X running Linux Mint 21.3

-- EXACT TEST --
fisher-rxc      0.2631  in 21.11s
rpy2            0.2631  in 26.15s
-----------
fisher-rxc      0.9981  in 4.95s
rpy2            0.9981  in 6.96s
-----------
-- MONTE-CARLO SIMULATION --
fisher-rxc      0.2631  in 0.60s
rpy2            0.2630  in 16.10s
-----------
fisher-rxc      0.9981  in 0.62s
rpy2            0.9981  in 19.38s
-----------

References

Contingency table generator (ASA159): https://people.sc.fsu.edu/~jburkardt/c_src/asa159/asa159.html

Fisher' exact test - network algorithm (ASA643): https://netlib.org/toms/643.gz

Fisher's exact test - recursive: https://stackoverflow.com/questions/25368284/fishers-exact-test-for-bigger-than-2-by-2-contingency-table

Fortran to C transpiler: https://www.netlib.org/f2c/

C to Rust transpiler: https://github.com/immunant/c2rust