Bayesian Network Inference Algorithms - Bayesian Networks in R: With Applications in Systems Biology - page 92

P ( pakts473 ) P ( PKA )

without evidence
with evidence

without evidence
with evidence

HIGH HIGH

AVG AVG

LOW LOW

0.0 0.2 0.4 0.6 0.2 0.4 0.6

probability probability

Fig. 4.1 Probability distributions of pakts473 and PKA before and after inhibiting p44.42

> querygrain(jprop, nodes = "PKA")$PKA
PKA

LOW AVG HIGH
0.48908954 0.45116629 0.05974417

All the queries illustrated above can be easily changed to maximum a posteriori
queries by finding the largest element in the distribution of the target node.

> names(which.max(querygrain(jprop,
+ nodes = c("PKA"))$PKA))

[1] "LOW"

4.2.2 Approximate Inference

The conditional probability queries from the previous section produce similar
results when they are performed with the logic sampling algorithm, illustrated
in Algorithm 4.2 and implemented in bnlearn in the cpdist and cpquery
functions.

> particles = cpdist(fitted, nodes = "pakts473",
+ evidence = (p44.42 == "LOW"))
> prop.table(table(particles))
particles
LOW AVG HIGH
0.665622103 0.332827458 0.001550438
> particles = cpdist(fitted, nodes = "PKA",
+ evidence = (p44.42 == "LOW"))

> prop.table(table(particles))
particles
LOW AVG HIGH
0.48865276 0.45108553 0.06026171