Computation of expected utility

andersrm · Post by **andersrm** » Tue Jan 27, 2009 4:15 am

Given an influence diagram that has a probabilistic node with several possible outcomes and an expected utility node as its child, how is expected utility calculated?

I would have expected: sum{j}{p_ j * u(x_ j) }, where the j denote the outcomes, the u(x_ j) denote the tabulated utilities for the j outcomes, and p_j are the computed probabilities.

Yet, in the attached file, GeNIe's computations of expected utility do not exactly match my computations for 2 of the 4 expected utility nodes that I have checked.

Does anyone know why this would be?

mark · Post by **mark** » Wed Jan 28, 2009 2:09 am

There must be an error in your calculations or a bug in GeNIe (or both). Can you please give an example where your calculations don't match the values computed by GeNIe?