#statistics #confidence #measurement #sets #level #compare #different

bin+lib nanostat

nanostat compares data sets using Welch's t-test at various levels of confidence

2 unstable releases

0.2.0 Dec 31, 2021
0.1.0 May 18, 2021

#10 in #confidence

MIT/Apache

17KB
180 lines

nanostat

Like ministat, but smaller?

A Rust library and CLI tool for evaluating whether two or more sets of measurements are statistically different. It does this by performing a Welch's t-test at a particular confidence level, making it suitable for small sets of measurements (e.g., multiple runs of a benchmark). It's inspired largely by FreeBSD's ministat (written by Poul-Henning Kamp).

$ nanostat examples/iguana examples/leopard examples/chameleon 
examples/leopard:
	Difference at 95% confidence!
		643.50 > 300.00 ± 293.97, p = .026
examples/chameleon:
	No difference at 95% confidence.

Install

cargo install nanostat --features cli

License

Copyright © 2021 Coda Hale

Distributed under the Apache License 2.0 or the MIT license.


lib.rs:

nanostat compares data sets using Welch's t-test at various levels of confidence.

Dependencies

~6.5MB
~127K SLoC