ignore/parallel-rng-notes.md

In TommyJones/tidylda: Latent Dirichlet Allocation Using 'tidyverse' Conventions

Here's what you need to do 1. Seed your RNG from R (still do this b/c C++ is faster) 2. Sample all your uniform or exponential RVs at once - consider doing this in parallel over how many threads you have - Only have to do the long jump thing once 3. Allocate those numbers in a structure that is cache access efficient - possibly best approach here is to have a nested vector of vectors of doubles of that conforms to [iterations][threads][doubles] 4. Rip through them as you sample

Another, possibly conceptually simpler thing to do 1. Seed the RNG once from R 2. For each iteration, sample threads number of integers 3. Pass those along to each thread to seed their RNG streams

TommyJones/tidylda documentation built on Jan. 16, 2025, 12:16 p.m.