Thanks for your question. We perform an analysis of how many samples are needed for convergence in the article's Python case study: github.com/fortitudo-te...
As we can see, something like 500 should be plenty sufficient, but we use 1000 as default because we also use it for other purposes.
