Today we learn that Rust’s Inverse Riffle Shuffle seems to be much (~10x) slower than their Fisher Yates shuffle on an [i32: 49].
The FY itself still isn’t “fast” - taking 1.6s to do 1M of them. Although if you swap to SmallRng instead of the plain old thread_rng, it drops to 0.6s/M.
Fun times.
Although through a combination of iteration, experimentation, frustration, randomly inserting and removing sigils, and, most crucially, thinking, I’ve got it down to 0.03s/M iterations. Which makes it a lot easier to gather more data for sampling.
Does seem to be biasing on the low side though - averaging about 12M runs for [1, 2, 3, 4, 5, 6] to appear in 4800 samples.
@zimpenfish hmmmm
If you’re counting or using the small end of numbers you might get that. Lower ordinal numbers appear more frequently for obvious reasons.
