study8nd: extracts sub-bn(s) with one parent one child from a bn
In rebastaba: Handling Bayesian networks

Description Usage Arguments Details Value Examples

(bn) From a given bn and for one specified node (ion), the list of all 'one relevant root' versus 'this child' bn(s) is constructed. Non considered parents are put at a fixed value.

1	study8nd(bn, ion, unfixed=character(0))

`bn`	The original bn.
`ion`	The definition of the target node (so-called child)
`unfixed`	The list of parents not to be fixed. This list can comprise other thing that parent names, they will be neglected.

For the sake of simplicity, let us call here as "parent" a root of the bn which has the studied node as descendant. The non varying parents are fixed with Dirac distributions for the continuous variables and single domain for categoric variables. The unique varying parent is inherited from a bn2bn call.
It must be underlined that the desired construction is not as straightforward as one can think at first. The difficulty comes with the fact that the parents are not necessarily independant. Let us consider the simple case: (A->C; B->C). C has got two parents which are independent and we probably be happy in studying the two probabilistic relations: [C|A,B=b] and [C|A=a,B] for respectively assess the influence of A over C; and of B over C. To do so we use these conditional probability adding a uniform distribution onto A and B
Now if we add a third arc: (A->B), this is not so clear because A and B are no longer independent! In that case, a choice could be to study sum\_b([C|A,B=b]) and [C|B,A=a] loosing the symetry between the two parents to follow the structure of the graph.
But things are indeed much more complicated since the joint probability of (A,B) can depend on non direct parents. Let us add a new node (D) and the arc (D->B)... Then natural proposals are not so spontaneous.
This is why the retained choice here, was to study the variation of one node for root ancestors. If somebody wants to use directly the direct parents, s/he must breaks the relationships between the parents down (assuming their independence); this can be done with bn2bn function.

A list of length the number of generated bn, each component comprising a bn. For each, description@comm[1] comprises the model associated to the bn, description@comm[2] comprises the list of fixed nodes.