fs.mrpval: Significance of Feature Ranking

In aberHRML/FIEmspro: Flow In-jection Electrospray Mass Spectrometry Processing: \\ data processing, classification modelling and variable selection in metabolite fingerprinting

Description

Computation of the pseudo mrp-value from a resampling based feature ranking strategy. qtl represents the fraction of presumedly informative features. The decision is based on the average rank across all resampling steps. 1-qtl represents the fraction of features that serves to estimate the null distribution of ranks (i.e. ranks of uninformative variables).

Usage

  fs.mrpval(x,qtl=0.75)

Arguments

x	A list returned from feat.rank.re.
qtl	A numeric value of probability with values in [0,1].

Value

A list with components:

stats	Original feature ranking statistics.
fs.rank	Feature ranking vector.
fs.order	Feature order vector.
sdrank	Feature rank standard deviation.
mrpval	Individual feature mrp-value.
Ug	Uninformative variables.
nnull	Total number of uninformative variables.
qtl	Quantile qtl used.

Author(s)

David Enot [e-mail] and Wanchang Lin [e-mail]

References

Zhang, C., Lu,X. and Zhang, X. (2006). Significance of Gene Ranking for Classification of Microarray Samples. IEEE/ACM Transactions on Computational Biology and Bioinformatics, VOL. 3, NO. 3, pp. 312-320.

See Also

feat.rank.re, fs.summary

Examples

## load abr1
data(abr1)
y   <- factor(abr1$fact$class)
x <- preproc(abr1$pos , y=y, method=c("log10","TICnorm"),add=1)[,110:500]  
## Select classes 1 and 2
dat <- dat.sel(x, y, choices=c("1","2"))
x <- dat$dat[[1]]
y <- dat$cl[[1]]

## partitioning
pars   <- valipars(sampling="boot",niter=2,nreps=5)
tr.idx <- trainind(y,pars=pars)

## multiple rankings using AUC
z      <- feat.rank.re(x,y,method="fs.auc",pars = pars,tr.idx=tr.idx)

## Compute stability mr-p value using the 25% worst features as irrelevant
res <- fs.mrpval(z,qtl=0.75)

## print content of res
names(res)

## list of features to form the null distribution of ranks
print(res$Ug)

aberHRML/FIEmspro documentation built on May 16, 2019, 6:56 p.m.