Talk:Bayesian network theory


 * odd feature of example 3: it is possible to have both a heart attack and nothing! Or a heart attack, stroke, and nothing.  Seems that merging these three child nodes into one may allow for more logical outcomes.  E.g. one child node with 4 states of: heart attack, stroke, nothing, heart attack & stroke.
 * the DBN description starts with a strange first step that has a different topology. Why is this?  As a note, it need not start this way, but could more simply just start with the core DBN.
 * for the DBN it may also be useful to show the rolled and unrolled structure. The last figure in the DBN section is an example of an "unrolled" structure.  Since this network has connections within the time slice (e.g. B3->C3, C3->A3, and B3->A3) it does not have a clean rolled structure.  If it were rolled it would look something like (B->A, B->B, C->A, B->A).  This rolled structure has loops and as such is not a DAG, but this lack of DAG structure is an artifact of how the rolled representation presents information.