decisionSupport (version 1.105.3)

make_CPT: Make Conditional Probability tables using the likelihood method

Description

This function creates Conditional Probability Tables for Bayesian Network nodes from parameters that (for complex nodes) can be more easily elicited from experts than the full table. The function uses the Likelihood method, as described by Sjoekvist S & Hansson F, 2013. Tables are created from three the relative weights of all parents, rankings for all parents, a parameter (b) for the sensitivity of the child node and a prior distribution (for the child node).

Usage

make_CPT(
  parent_effects,
  parent_weights,
  b,
  child_prior,
  ranking_child = NULL,
  child_states = NULL,
  parent_names = NULL,
  parent_states = NULL
)

Arguments

parent_effects

list of vectors describing the effects of all parent node states on the value of the child variable. For example, if parent 1 has four states, the respective vector might look like this: c(3,1,0,0). This would imply that the first state of the parent is strongly associated with high values for the child, the second less strongly, and the 3rd and 4th value are associated with equally low values.

parent_weights

weight factors for the parent nodes

b

parameter for the strength of the parent's influence on the child node. A value of 1 causes no response; 3 is quite strong.

child_prior

prior distribution for the states of the child node.

ranking_child

vector of length length(child_prior) containing rankings for the child node states on a -1..1 scale. If this is null, evenly spaced rankings on this -1..1 scale are assigned automatically.

child_states

optional vector specifying the names of the child states.

parent_names

optional vector specifying parent node names.

parent_states

list of the same structure as parent_effects containing names for all states of all parents.

Value

list of two data.frames: 1) Conditional Probability Table (CPT); 2) legend table specifying which states of the parent nodes belong to which column in the CPT.

References

Sjoekvist S & Hansson F, 2013. Modelling expert judgement into a Bayesian Belief Network - a method for consistent and robust determination of conditional probability tables. Master's thesis, Faculty of Engineering, Lund University; http://lup.lub.lu.se/luur/download?func=downloadFile&recordOId=3866733&fileOId=3866740

Examples

# NOT RUN {
make_CPT(parent_effects=list(c(-1,1),c(-0.5,0,0.5)),
  parent_weights=c(3,1),b=1.5,child_prior=c(.2,.6,.2),child_states=c("a","b","c"))

test_CPT<-make_CPT(parent_effects=list(c(-1,3),c(-4,2),c(-2,3,4),c(1,2,3)),
                  parent_weights=c(1,1,1,1),b=2,child_prior=c(1,2,3,4,5),
                  child_states=c("a","b","c","d","e"),
                  parent_states=list(c("low","high"),c("A","B"),c(1,2,3),c("Hi","Lunch","Bye")))


# }

Run the code above in your browser using DataCamp Workspace