tpc: PC Algorithm Accounting for a Partial Node Ordering

In tpc: Tiered PC Algorithm

PC Algorithm Accounting for a Partial Node Ordering

Description

Like [pcalg::pc()], but takes into account a user-specified partial ordering of the nodes/variables. This has two effects: 1) The conditional independence between x and y given S is ot tested if any variable in S lies in the future of both x and y; 2) edges cannot be oriented from a higher-order to a lower-order node. In addition, the user may specify individual forbidden edges and context variables.

Usage

tpc(
  suffStat,
  indepTest,
  alpha,
  labels,
  p,
  skel.method = c("stable", "stable.parallel"),
  forbEdges = NULL,
  m.max = Inf,
  conservative = FALSE,
  maj.rule = TRUE,
  tiers = NULL,
  context.all = NULL,
  context.tier = NULL,
  verbose = FALSE,
  numCores = NULL,
  cl.type = "PSOCK",
  clusterexport = NULL
)

Arguments

suffStat	A [base::list()] of sufficient statistics, containing all necessary elements for the conditional independence decisions in the function [indepTest()].
indepTest	A function for testing conditional independence. It is internally called as indepTest(x,y,S,suffStat), and tests conditional independence of x and y given S. Here, x and y are variables, and S is a (possibly empty) vector of variables (all variables are denoted by their (integer) column positions in the adjacency matrix). suffStat is a list, see the argument above. The return value of indepTest is the p-value of the test for conditional independence.
alpha	significance level (number in (0,1) for the individual conditional independence tests.
labels	(optional) character vector of variable (or "node") names. Typically preferred to specifying p.
p	(optional) number of variables (or nodes). May be specified if labels are not, in which case labels is set to 1:p.
skel.method	Character string specifying method; the default, "stable" provides an order-independent skeleton, see [tpc::tskeleton()].
forbEdges	A logical matrix of dimension p*p. If [i,j] is TRUE, then the directed edge i->j is forbidden. If both [i,j] and [j,i] are TRUE, then any type of edge between i and j is forbidden.
m.max	Maximal size of the conditioning sets that are considered in the conditional independence tests.
conservative	Logical indicating if conservative PC should be used. Defaults to FALSE. See [pcalg::pc()] for details.
maj.rule	Logical indicating if the majority rule should be used. Defaults to TRUE. See [pcalg::pc()] for details.
tiers	Numeric vector specifying the tier / time point for each variable. Must be of length 'p', if specified, or have the same length as 'labels', if specified. A smaller number corresponds to an earlier tier / time point.
context.all	Numeric or character vector. Specifies the positions or names of global context variables. Global context variables have no incoming edges, i.e. no parents, and are themselves parents of all non-context variables in the graph.
context.tier	Numeric or character vector. Specifies the positions or names of tier-specific context variables. Tier-specific context variables have no incoming edges, i.e. no parents, and are themselves parents of all non-context variables in the same tier.
verbose	if TRUE, detailed output is provided.
numCores	The numbers of CPU cores to be used.
cl.type	The cluster type. Default value is "PSOCK". For High-performance clusters use "MPI". See also parallel::makeCluster.
clusterexport	Character vector. Lists functions to be exported to nodes if numCores > 1.

Details

See pcalg::pc for further information on the PC algorithm. The PC algorithm is named after its developers Peter Spirtes and Clark Glymour (Spirtes et al., 2000).

Specifying a tier for each variable using the tier argument has the following effects: 1) In the skeleton phase and v-structure learing phases, conditional independence testing is restricted such that if x is in tier t(x) and y is in t(y), only those variables are allowed in the conditioning set whose tier is not larger than t(x). 2) Following the v-structure phase, all edges that were found between two tiers are directed into the direction of the higher-order tier. If context variables are specified using context.all and/or context.tier, the corresponding orientations are added in this step.

Value

An object of class "pcAlgo" (see [pcalg::pcalgo] containing an estimate of the equivalence class of the underlying DAG.

Author(s)

Original code by Markus Kalisch, Martin Maechler, and Diego Colombo. Modifications by Janine Witte (Kalisch et al., 2012).

References

M. Kalisch, M. Maechler, D. Colombo, M.H. Maathuis and P. Buehlmann (2012). Causal Inference Using Graphical Models with the R Package pcalg. Journal of Statistical Software 47(11): 1–26.

P. Spirtes, C. Glymour and R. Scheines (2000). Causation, Prediction, and Search, 2nd edition. The MIT Press. https://philarchive.org/archive/SPICPA-2.

Examples

# load simulated cohort data
data(dat_sim)
n <- nrow(dat_sim)
lab <- colnames(dat_sim)

# estimate skeleton without taking background information into account
tpc.fit <- tpc(suffStat = list(C = cor(dat_sim), n = n),
               indepTest = gaussCItest, alpha = 0.01, labels = lab)
pc.fit <- pcalg::pc(suffStat = list(C = cor(dat_sim), n = n),
                    indepTest = gaussCItest, alpha = 0.01, labels = lab,
                    maj.rule = TRUE, solve.conf = TRUE)
identical(pc.fit@graph, tpc.fit@graph) # TRUE
# estimate skeleton with temporal ordering as background information
tiers <- rep(c(1,2,3), times=c(3,3,3))
tpc.fit2 <- tpc(suffStat = list(C = cor(dat_sim), n = n),
                indepTest = gaussCItest, alpha = 0.01, labels = lab, tiers = tiers)

tpc.fit3 <- tpc(suffStat = list(C = cor(dat_sim), n = n),
                indepTest = gaussCItest, alpha = 0.01, labels = lab, tiers = tiers,
                skel.method = "stable.parallel",
                numCores = 2, clusterexport = c("cor", "ecdf"))

if(requireNamespace("Rgraphviz", quietly = TRUE)){
 data("true_sim")
 oldpar <- par(mfrow = c(1,3))
 plot(true_sim, main = "True DAG")
 plot(tpc.fit, main = "PC estimate")
 plot(tpc.fit2, main = "tPC estimate")
 par(oldpar)
 }

 # require that there is no edge between A1 and A1, and that any edge between A2 and B2
 # or A2 and C2 is directed away from A2
 forb <- matrix(FALSE, nrow=9, ncol=9)
 rownames(forb) <- colnames(forb) <- lab
 forb["A1","A3"] <- forb["A3","A1"] <- TRUE
 forb["B2","A2"] <- TRUE
 forb["C2","A2"] <- TRUE

 tpc.fit3 <- tpc(suffStat = list(C = cor(dat_sim), n = n),
                 indepTest = gaussCItest, alpha = 0.01,labels = lab,
                 forbEdges = forb, tiers = tiers)

 if (requireNamespace("Rgraphviz", quietly = TRUE)) {
 # compare estimated CPDAGs
   data("true_sim")
   oldpar <- par(mfrow = c(1,2))
   plot(tpc.fit2, main = "old tPC estimate")
   plot(tpc.fit3, main = "new tPC estimate")
   par(oldpar)
 }
 # force edge from A1 to all other nodes measured at time 1
 # into the graph (note that the edge from A1 to A2 is then
 # forbidden)
 tpc.fit4 <- tpc(suffStat = list(C = cor(dat_sim), n = n),
                 indepTest = gaussCItest, alpha = 0.01, labels = lab,
                 tiers = tiers, context.tier = "A1")

 if (requireNamespace("Rgraphviz", quietly = TRUE)) {
 # compare estimated CPDAGs
  data("true_sim")
  plot(tpc.fit4, main = "alternative tPC estimate")
 }

 # force edge from A1 to all other nodes into the graph
 tpc.fit5 <- tpc(suffStat = list(C = cor(dat_sim), n = n),
                 indepTest = gaussCItest, alpha = 0.01, labels = lab,
                 tiers = tiers, context.all = "A1")

 if (requireNamespace("Rgraphviz", quietly = TRUE)) {
 # compare estimated CPDAGs
 data("true_sim")
 plot(tpc.fit5, main = "alternative tPC estimate")
 }

tpc documentation built on March 7, 2023, 7:18 p.m.