Note that we can use the @functools.lru_cache(None) to easily benefit from memoization.