For a better (but much slower) approximation, increase the number of samples generated to 1,000,000.