Not all available observations matter when a decision must be made. Intuitively, a parent of a decision node is said to be requisite if the value of the parent may affect the optimal choice of the decision.
The performance of inference in a LIMID can be improved, if the network is simplified by removing the nonrequisite parents of all decision nodes.
Lauritzen and Nilsson (2001) present an algorithm for removing the nonrequisite parents of the decision nodes in a LIMID. The result of this algorithm is a LIMID, where no parent of a decision node is nonrequisite. This network is known as the minimal reduction of the LIMID.
Our definition of requisiteness is slightly different than the one given by Lauritzen and Nilsson: We define a parent of a decision node to be requisite if and only if it is also a parent of the decision node in the minimal reduction of the LIMID.
Inspect the requisite parents of a decision node¶
Select a decision node. Click Network -> Analysis -> Show requisite parents for decision.
Requisite parents will be marked with a green V (see Figure 1) and non-requisite parents will be marked with a red X (see Figure 2).
If the model is a class, this function is only available in run-mode. If the model is a flat net, the function is available in both edit-mode and run-mode.