Approximate randomization tests

[Daniël de Kok]

Hi all,

To test the significance in the evaluation of (amongst other things) parse disambiguation and fluency ranking, I developed a module and two utilities for performing approximate randomization tests:

https://github.com/danieldk/approx-rand-test

Currently, it supports both paired and non-paired tests and various test statistics. Since I plan to submit this for the statistics package, I am hesitant to put the package to Hackage. However, there are currently two blockers:

- The statistics package does not return p-values for tests [1]. So, we currently use our own TestResult data type, which does return the p-value.
- The statistics package relies on mwc-random as a source of random numbers. However, in my tests, mersenne-random-pure64 was at least twice faster. So, my package currently uses mersenne-random-pure64 and monad-mersenne-random. (I sent a test case to Bryan O'Sullivan.)

Take care,
Daniël de Kok

[1] https://github.com/bos/statistics/issues/25