grain_jevidence: Set joint evidence in grain objects

In DataSciBurgoon/gRain: Graphical Independence Networks

Description

Setting and removing joint evidence in grain objects.

Usage

setJEvidence(object, evidence = NULL, propagate = TRUE, details = 0)

retractJEvidence(object, items = NULL, propagate = TRUE, details = 0)

new_jev(ev, levels)

Arguments

object	A "grain" object.
evidence	A list of evidence. Each element is a named array.
propagate	Should evidence be absorbed once entered; defaults to TRUE.
details	Amount of printing; for debugging.
items	Items in the evidence list to be removed. Here, NULL means remove everything, 0 means nothing is removed. Otherwise items is a numeric vector.
ev	A named list.
levels	A named list.

Note

All the joint evidence functionality should be used with great care.

Author(s)

Søren Højsgaard, [e-mail]

Examples

data(chest_cpt)
chest.bn <- grain(compileCPT(chest_cpt))
chest.bn <- compile(chest.bn)

uni <- list(asia = c("yes", "no"), tub = c("yes", "no"),
            smoke = c("yes", "no"), lung = c("yes", "no"),
            bronc = c("yes", "no"), either = c("yes", "no"),
            xray = c("yes", "no"), dysp = c("yes", "no"))

ev <- list(tab("asia", levels=uni, values=c(1,0)),
           tab("dysp", levels=uni, values=c(1,0)),
           tab(c("dysp","bronc"), levels=uni, values=c(.1, .2, .9, .8)) )

chest.bn
chest.bn2 <- setJEvidence(chest.bn, evidence=ev)
chest.bn2

# Notice: The evidence is defined on (subsets of) cliques of the junction tree
# and therefore evidence can readily be absorbed:
getgrain(chest.bn, "rip")$cliques  %>% str

# On the other hand, below is evidence which is not defined cliques
# of the junction tree and therefore evidence can not easily be
# absorbed.  Hence this will fail:

## Not run: 
ev.fail <- list(tab(c("dysp","smoke"), levels=uni, values=c(.1, .2, .9, .8)) )
setJEvidence(chest.bn, evidence=ev.fail)

## End(Not run)

## Evidence can be removed with

retractJEvidence(chest.bn2)      ## All evidence removed.
retractJEvidence(chest.bn2, 0)   ## No evidence removed.
retractJEvidence(chest.bn2, 1:2) ## Evidence items 1 and 2 are removed.

# Setting additional joint evidence to an object where joint
# evidence already is set will cause an error. Hence this will fail:

## Not run: 
  ev2 <- list(smoke="yes")
  setJEvidence(chest.bn2, evidence=ev2)

## End(Not run)

## Instead we can do
new.ev <- c(getEvidence(chest.bn2), list(smoke="yes"))
chest.bn
setJEvidence(chest.bn, evidence=new.ev)

## Create joint evidence object:
yn <- c("yes", "no")
db <- parray(c("dysp", "bronc"), list(yn, yn), values=c(.1, .2, .9, .8))
db
ev   <- list(asia=c(1, 0), dysp="yes", db)

jevi <- new_jev(ev, levels=uni)
jevi

chest.bn3 <- setJEvidence(chest.bn, evidence=jevi)
evidence(chest.bn3)

DataSciBurgoon/gRain documentation built on March 25, 2020, 12:02 a.m.