Talk:Spectral test

Source for 2nd and 3rd sentences
The source I used for the description of how the test works is from a lecture at a university. The manual mentions nothing about citing lectures or presentations by professors, so I assume that they are okay, so long as they are accurate. That description also matches with my personal understanding of how the test works, which is the other reason I used it. Still, if somebody finds a different source on the same topic that may be more reliable, I would be grateful if you replaced mine with it. — Preceding unsigned comment added by Skylord a52 (talk • contribs) 01:54, 2 December 2016 (UTC)

I know this test is dumb enough to be noteworthy how dumb it is
I can see how a LCG that only returns odd values is 100% wasteful on the least significant bit. But this test seems to be more specific than being on that general LCG flaw on lower significant bits having a shorter period than higher significant bits.

But failing to address that in a general form. And that makes it dumb.

Showing all possible outputs in a lattice over n domains, calculating some distance or lines or hyper-planes is a measure, but it seems quite nonsensical and irrelevant, because the node traversal is not even shown in that lattice projection. In comparison, i can draw a 3d Hilbert curve 2d projected, only in dots, and say "wow, that Hilbert curve is just dots at projected cube corners". But I am not dumb enough to write that without satire, excluding Poe's law.

The lowest example image shows 2 lattices, that are just rotated and stretched differently. They just appear different due to the modulo operator, but when you stack multiple of them in a tiling to undo that, they are the same lattices with a different ratio. the remaining argument is: "(closer to) golden ratio lattice is likely a better lattice than (closer to) square lattice, if only for more even distribution"

then is dumb enough to add some arbitrary bias to it, that has as much weight as "the apple on the right is clearly obviously objectively and arbitrarily axiomatically a worse fruit than the pea on the left" this test is also always biased in how it folds its domains, as different folding of domains will result in different distances of hyper-planes. this is just as silly as favoring the folding of one hyperbolic Hilbert curve over another, in blissful ignorance of arguing about bias and semantics, but not about content.

At least as this test is described here, in its basics, it is just dumb. There may be a "hidden gem in this pile of shit", but I doubt it.

Ollj (talk) 18:36, 3 September 2017 (UTC) User:Ollj


 * I recognize that the source I used for the bottom image may not be the best, and I would welcome a replacement if you have any. As for your opinion on the rest of the article, feel free to edit it to include your concerns about the test's effectiveness, if you can find a reputable source that backs them up. Keep in mind, though, that the test was vouched for by Knuth himself, which is a pretty high bar to overcome in terms of credibility.


 * Also, there's generally no need for that level of rudeness and vulgarity. Try to keep any issues you have with the subject of the article separate from those you have with the article itself.
 * Skylord a52 (talk) 21:17, 15 July 2018 (UTC)