You're right. I had a bug in my KL function, where I let some log(0)'s = 0. Thanks --good catch! I propose we calculate the KL's how you suggested. Do others agree?