The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.

Basic usage of grpSLOPE

Alexej Gossmann

2023-05-23

Introduction

Group SLOPE (gSLOPE) is a penalized linear regression method that is used for adaptive selection of groups of significant predictors in a high-dimensional linear model. A unique property of the Group SLOPE method is that it offers group false discovery rate (gFDR) control (i.e., control of the expected proportion of irrelevant groups among the total number of groups selected by Group SLOPE). A detailed description of the method can be found in D. Brzyski, A. Gossmann, W. Su, and M. Bogdan (2016) Group SLOPE — adaptive selection of groups of predictors.

Group SLOPE is implemented in the R package grpSLOPE. As an introduction to the R package, in the following we will walk through a basic usage demonstration. First, we will simulate some data, before we feed it into grpSLOPE, and subsequently examine the output.

Data generation

We simulate a \(500 \times 500\) SNP-data-like model matrix.

set.seed(17082016)

p     <- 500
probs <- runif(p, 0.1, 0.5)
probs <- t(probs) %x% matrix(1,p,2)
X0    <- matrix(rbinom(2*p*p, 1, probs), p, 2*p)
X     <- X0 %*% (diag(p) %x% matrix(1,2,1))

For example, the upper left \(10 \times 10\) corner of \(X\) looks as follows.

1 1 0 1 0 0 1 2 0 0
0 0 1 1 0 0 1 1 1 0
1 0 1 1 0 0 1 0 0 0
1 0 0 0 1 0 0 0 2 2
1 0 0 1 1 1 0 0 1 1
0 1 0 0 0 0 2 1 2 2
0 0 0 0 1 2 0 0 2 0
2 0 0 0 0 0 0 1 0 2
1 0 1 0 0 1 0 1 1 1
0 0 1 0 0 0 0 0 1 1

Note: In fact, with the default settings, the Group SLOPE method is guaranteed to control gFDR only when applied to a data matrix, where the columns corresponding to different groups of predictors are nearly uncorrelated. The relevant theoretical results can be found in Brzyski et. al. (2016). Only for the brevity of exposition we neither check for nor enforce low between-group correlations in this example.

We divide the 500 predictor variables into 100 groups of sizes ranging from 3 to 7.

group <- c(rep(1:20, each=3),
           rep(21:40, each=4),
           rep(41:60, each=5),
           rep(61:80, each=6),
           rep(81:100, each=7))
group <- paste0("grp", group)
str(group)
##  chr [1:500] "grp1" "grp1" "grp1" "grp2" "grp2" "grp2" "grp3" "grp3" "grp3" ...

For further usage we keep additional information about the grouping structure of predictors, such as the total number of groups and the group sizes.

# this generates a list containing a vector of indices for each group:
group.id <- grpSLOPE::getGroupID(group)
# this extracts the total number of groups:
n.group <- length(group.id)
# this vector collects the sizes of every group of predictors:
group.length <- sapply(group.id, FUN=length)
# this vector collects the group names:
group.names <- names(group.id)

In order to simulate a response variable, we randomly select 10 groups to be truly significant.

ind.relevant <- sort(sample(1:n.group, 10)) # indices of relevant groups

The randomly selected truly significant groups are:

grp4 grp12 grp25 grp34 grp35 grp38 grp66 grp67 grp81 grp89

Then we generate the vector of regression coefficients, by sampling effect sizes for the significant groups from the Uniform(0,1) distribution.

b <- rep(0, p)
for (j in ind.relevant) {
  b[group.id[[j]]] <- runif(group.length[j])
}

Finally, we generate the response vector according to a linear model with i.i.d. \(\mathcal{N}(0, 1)\) noise terms.

y <- X %*% b + rnorm(p)

Fitting the Group SLOPE model

We fit the Group SLOPE model to the simulated data. The function argument fdr signifies the target group-wise false discovery rate (gFDR) of the variable selection procedure.

library(grpSLOPE)

model <- grpSLOPE(X=X, y=y, group=group, fdr=0.1)

Model fit results

The resulting object model of class “grpSLOPE” contains a lot of information about the resulting Group SLOPE model. Some of these parameters are shown below.

We can further check the performance of the method by computing the resulting group false discovery proportion (gFDP) and power.

true.relevant <- group.names[ind.relevant]
truepos       <- intersect(model$selected, true.relevant)

n.truepos  <- length(truepos)
n.selected <- length(model$selected)
n.falsepos <- n.selected - n.truepos

gFDP <- n.falsepos / max(1, n.selected)
pow <- n.truepos / length(true.relevant)

print(paste("gFDP =", gFDP))

[1] “gFDP = 0.230769230769231”

print(paste("Power =", pow))

[1] “Power = 1”

We see that the method indeed did not exceed the target gFDR, while maintaining a high power.

Lambda sequences

Multiple ways to select the regularizing sequence \(\lambda\) are available.

If a group structure with little correlation between groups can be assumed (i.e., groups in the standardized model matrix are nearly orthogonal), then we suggest to use the sequence “corrected”, which is the default.

The \(\lambda\) sequences “mean” and “max” can be used together with the options orthogonalize = FALSE and normalize = FALSE, when the columns of the model matrix are exactly orthogonal to each other (“max” is more conservative, giving exact gFDR control only when all groups have the same size, and otherwise resulting in a lower gFDR than the target level).

Alternatively, any non-increasing sequence of appropriate length can be utilized. However, we do not recommend to use any other \(\lambda\) sequences unless you really know what you are doing.

References

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.