Biology Reference
In-Depth Information
> dag3 = empty.graph(nodes(dag))
> dag3 = set.arc(dag3, "VECT", "MECH")
> dag3 = set.arc(dag3, "ALG", "MECH")
> dag3 = set.arc(dag3, "ALG", "VECT")
> dag3 = set.arc(dag3, "ANL", "ALG")
> dag3 = set.arc(dag3, "STAT", "ALG")
> dag3 = set.arc(dag3, "STAT", "ANL")
> all.equal(dag, dag3)
[1] TRUE
The approaches discussed above are guaranteed to result in directed or partially
DAGs unless
check.cycles
is explicitly set to
FALSE
. A quick check reveals
that the moral graph of
dag
and the graphical model from
Whittaker
(
1990
) express
the same dependence relationships, as expected.
> all.equal(ug, moral(dag))
[1] TRUE
Upon creating a
bn
object, we are in a position to investigate those properties of
the corresponding graph that have a probabilistic interpretation in a Bayesian net-
work. For this purpose, the
bn
class provides a complete description of the network
structure (which is uniquely specified by its arc set), and the use of the information
stored for each node results in significant performance improvements for common
operations.
For instance, when treating the network as a causal model we are often interested
in the topological ordering of the nodes. The relative position of two nodes in the
topological ordering is indicative of the direction of any possible causal relationship
between them, because it implies the direction of any possible path linking the nodes
(a more detailed explanation can be found in Sect.
2.4
).
> node.ordering(dag)
[1] "STAT" "ANL" "ALG" "VECT" "MECH"
The neighborhood (
nbr
) and the Markov blanket (
mb
) of a node provide a syn-
thetic description of the local dependence structure around that node. These can be
obtained as follows:
> nbr(dag, "ANL")
[1] "ALG" "STAT"
> mb(dag, "ANL")
[1] "ALG" "STAT"
We can also use the commands above to show that both sets describe symmetric
relationships, i.e., if
ALG
is in the Markov blanket of
ANL
,
ANL
is in the Markov
blanket of
ALG
.
> "ANL" %in% mb(dag, "ALG")
[1] TRUE
> "ALG" %in% mb(dag, "ANL")
[1] TRUE
Search WWH ::
Custom Search