diff options
Diffstat (limited to 'pelican/content/RandomnessTesting.md')
| -rw-r--r-- | pelican/content/RandomnessTesting.md | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/pelican/content/RandomnessTesting.md b/pelican/content/RandomnessTesting.md index 4272bd9..67d528e 100644 --- a/pelican/content/RandomnessTesting.md +++ b/pelican/content/RandomnessTesting.md @@ -41,11 +41,11 @@ The way `dieharder` works, it simply returns a clear assessment of the test resu ### Caveats There are a number of things to keep in mind when using `dieharder`, especially so when running it on a reduced amount of test data. - * If `dieharder` reaches the end of the input file, it rewinds and uses the same test data again. This has rather drastic effects on several tests which assume that some sort of repetition in the input is a sign for a seriously flawed generator. - * The `-Y 1` option works by adding 100 to the value of `psamples` until a conclusive result is found. This works reasonably well with tests that start with a value of 100 to `psamples`, but there are tests starting with 1000, and others starting with 1. - * The `-m <n>` option also just affects the initial value of `psamples`. - * Some of the tests themselves are marked as "suspect" or "do not use" if you run `dieharder -l`. Still, `-a` runs them, for whatever reason. - * Expect about one test in a `-a` run to return a "weak" result that needs to be resolved. According to the man page, about 1 in 1000 tests (not `-a` runs!) may fail despite the fact that the input is good. In that case I suggest doubling the input size, either by using `-m <n>` (which only works in conjunction with `-a`) or by adjusting `psamples` and/or `tsamples` manually. +* If `dieharder` reaches the end of the input file, it rewinds and uses the same test data again. This has rather drastic effects on several tests which assume that some sort of repetition in the input is a sign for a seriously flawed generator. +* The `-Y 1` option works by adding 100 to the value of `psamples` until a conclusive result is found. This works reasonably well with tests that start with a value of 100 to `psamples`, but there are tests starting with 1000, and others starting with 1. +* The `-m <n>` option also just affects the initial value of `psamples`. +* Some of the tests themselves are marked as "suspect" or "do not use" if you run `dieharder -l`. Still, `-a` runs them, for whatever reason. +* Expect about one test in a `-a` run to return a "weak" result that needs to be resolved. According to the man page, about 1 in 1000 tests (not `-a` runs!) may fail despite the fact that the input is good. In that case I suggest doubling the input size, either by using `-m <n>` (which only works in conjunction with `-a`) or by adjusting `psamples` and/or `tsamples` manually. ### Installation |