On computing throttling rate limits in web APIs through statistical inference