How does JMP bias the rate of count when calculating the probability of Partition platform with categorical responses?

[#1]

For categorical responses Partition has the response rates that are slightly biased away from zero to avoid having attributed probabilities of zero. I would like to know how to calcurate the biased probabilities.