Ordering of branches in TreeADDs in consistency with the variable domain

[Manuel Luque]

I find the ordering of branches in TreeADDs is not completely consistent with the ordering of states in the variable of the potential being edited.

For example, let us consider the attached Bayesian network where variable B has four states: absent, mild, moderate, and severe. (Let us note that it is a standard domain of OpenMarkov, with states in increasing order of severity). Let us assume that we edit the probability potential of variable A and we want to assign a TreeADD.

If we start creating the TreeADD, then we would have something as in the next figure. The least severe state appears below ("absent"), and the states appear in increasing order of severity.

Let us now assume that we add states to the first branch, labeled as "severe", and the states added are "moderate" and "mild". Then, we would have something as in the next figure.

And now let us assume now that the user reconsiders the organization of branches and then she/he removes the states "moderate" and "mild" from the first branch. Then, we would obtain something as in the next figure.

In the last figure it can be seen that the order is: absent, severe, {mild, moderate}, which means they are not in a consistent order of severity (upwards or downwards).

In my opinion, the order of branches, when possible, should be consistent with the order of states in the domain of the variable. For example, in last figure the branches should be in the next order: severe, {mild, moderate}, absent