Bayesian Networks in the Absence of Temporal Information - Bayesian Networks in R: With Applications in Systems Biology - page 39

> bn.hc

Bayesian network learned via Score-based methods

model:
[MECH][VECT|MECH][ALG|MECH:VECT][ANL|ALG]
[STAT|ALG:ANL]
nodes: 5

arcs: 6

undirected arcs: 0

directed arcs: 6

average markov blanket size: 2.40

average neighbourhood size: 2.40

average branching factor: 1.20

learning algorithm:

Hill-Climbing

score:

Bayesian Information Criterion (Gaussian)
penalization coefficient: 2.238668
tests used in the learning procedure: 34
optimized: TRUE

This new network fits the marks data as well as the previous one, as they have the
same BIC score.

> score(bn.gs, data = marks, type = "bic-g")
[1] -1720.15
> score(bn.hc, data = marks, type = "bic-g")
[1] -1720.15
Even though the directions of the arcs are different, the arcs themselves are the same
in all the Bayesian networks we learned from the data. Furthermore, they are also the
same (again modulo their direction) as the ones present in the original model from
Whittaker ( 1990 ), which seems to indicate that the network structure we learned is
reliable.
Learning the structure of a network with deal requires a different workflow due
to the Bayesian approach to model selection adopted by the package, even though
it leads to the same result. First, we must define an empty network structure for
the data and fit the prior distribution, which has the form described in Bøttcher and
Dethlefsen ( 2003 ).

> net = network(marks)
> prior = jointprior(net, N = 5)
The argument N passed to jointprior is the imaginary or equivalent sample
size , which expresses the weight assigned to the prior distribution as the size of
an imaginary sample size supporting it. According to the experiments performed