I am slowly beginning to realize why one cannot learn parameters in an ID. However, my intuition tells me that it would make sense to learn parameters in the "corresponding BBN". By that I mean the BBN that will arise if I remove all the utility nodes from my ID and exchange all decision nodes with chance nodes that have the same states as their corresponding decision nodes.
My question is twofold: First, can someone confirm that I am right or wrong in my intuition? And second, if I am right, has anyone thought of adding that feature to GeNIe? (I.e. that if one tries to learn parameters in an ID GeNIe will assume that one wants to learn them in the "corresponding BBN".)
A suggestion for the help files and tutorials would be to add a few sentences that explain why parameter learning is disabled in the IDs.