ddgraph (version 1.16.0)

ncpc: Make a Direct Dependence Graph using the NCPC algorithm...

Description

Make a Direct Dependence Graph using the NCPC algorithm

Usage

ncpc(obj, alpha=0.05, p.value.adjust.method="none", test.type=c("mc-x2-c", "cor"), max.set.size=NULL, mc.replicates=5000, report.file=NULL, verbose=FALSE, star=FALSE, min.table.size=10)

Arguments

obj

DDDataSet object

alpha

the alpha (P-value) cutoff for conditional independence tests

p.value.adjust.method

the multiple testing correction adjustment method

test.type

the type of conditional independence test (default: Monte Carlo x2 test "mc-x2-c" for binary data and partial correlation "cor" for continuous data) . See the documentation for ciTest for other available conditional independence tests

max.set.size

the maximal number of variables to condition on, if NULL estimated from number of positives in class labels (default: NULL)

mc.replicates

the number of Monte-carlo replicates, if applicable (default: 5000)

report.file

name of the file where a detailed report is to be printed, reporting is suppressed if NULL (default: NULL)

verbose

if to print out information about how the algorithm is progressing (default: TRUE)

star

if to use the NCPC* algorithm (default: FALSE)

min.table.size

the minimal number of samples in a contingency table per conditioning set (makes sense only for discrete data)

Value

Details

Make a Direct Dependence Graph using a P-value and conditional independence tests. There are two version of the algorithm: NCPC and NCPC*. NCPC finds the causal neighbourhood while the NCPC* infers the full Markov Blanket.

The full algorithm is given in (Stojnic et al, 2012).

References

R. Stojnic et al (2012): "A Graphical Modelling Approach to the Dissection of Highly Correlated Transcription Factor Binding Site Profiles", in press, PloS Computational Biology.

Examples

data(mesoBin)
# run the NCPC algorithm with alpha=0.05 (on discrete data)
ncpc(mesoBin$Meso, alpha=0.05, test.type="mc-x2-c")
# run the NCPC* algorithm with alpha=0.05 (on discrete data)
res <- ncpc(mesoBin$Meso, alpha=0.05, test.type="mc-x2-c", star=TRUE)

# analysis of results:
class(res)
# although of class DDGraph, behaves much like a list
names(res)
# parameters used in obtaining results
res$params
# labels for each of the variables
res$final.calls
# direct variables
res$direct

### load continous data
data(mesoCont)
# run the NCPC algorith with alpha=0.05 (on continuous data)
ncpc(mesoCont$Meso, alpha=0.05, test.type="cor", max.set.size=1)
# run the NCPC* algorith with alpha=0.05 (on continuous data)
ncpc(mesoCont$Meso, alpha=0.05, test.type="cor", max.set.size=1, star=TRUE)