Hello everybody. I need to generate random discrete multinomial probability distributions of a big size (20 probability values each, more or less), and these generated distributions must be uniformly distributed. Simple methods like generating random numbers and normalizing after do not work well, as they introduce a big bias. I found just this one paper about the problem
where the authors show that their algorithm works for distributions with three values (Figure 3, last page), but I have tested it for bigger sizes (20, as mentioned above) and a big bias is introduced.
¿Does anyone know the state-of-the-art or some work about this problem?