Boost step by step functions

Step by step functions to apply the selectboost algorithm.

boost.normalize(X, eps = 1e-08)

boost.compcorrs(
  Xnorm,
  corrfunc = "cor",
  verbose = FALSE,
  testvarindic = rep(TRUE, ncol(Xnorm))
)

boost.correlation_sign(Correlation_matrice, verbose = FALSE)

boost.findgroups(Correlation_matrice, group, corr = 1, verbose = FALSE)

boost.Xpass(nrowX, ncolX)

boost.adjust(
  X,
  groups,
  Correlation_sign,
  Xpass = boost.Xpass(nrowX, ncolX),
  verbose = FALSE,
  use.parallel = FALSE,
  ncores = 4
)

boost.random(
  X,
  Xpass,
  vmf.params,
  verbose = FALSE,
  B = 100,
  use.parallel = FALSE,
  ncores = 4
)

boost.apply(
  X,
  cols.simul,
  Y,
  func,
  verbose = FALSE,
  use.parallel = FALSE,
  ncores = 4,
  ...
)

boost.select(Boost.coeffs, eps = 10^(-4), version = "lars", verbose = FALSE)

Arguments

X: Numerical matrix. Matrix of the variables.
eps: Numerical value. Response vector.
Xnorm: Numerical matrix. Needs to be centered and l2 normalized.
corrfunc: Character value or function. The function to compute associations between the variables.
verbose: Boolean. Defaults to FALSE.
testvarindic: Boolean vector. Compute associations for a subset of variables. By default, the scope of the computation is the whole dataset, i.e. rep(TRUE,ncol(Xnorm)).
Correlation_matrice: Numerical matrix.
group: Character value or function. The grouping function.
corr: Numerical value. Thresholding value. Defaults to 1.
nrowX: Numerical value
ncolX: Numerical value.
groups: List. List of groups or communities (compact form).
Correlation_sign: Numerical -1/1 matrix.
Xpass: Numerical value. Transformation matrix. Defaults to boost.Xpass(nrowX,ncolX), with nrowX=nrow(X) and ncolX=ncol(X).
use.parallel: Boolean. Defaults to FALSE.
ncores: Numerical value. Number of cores to use. Defaults to 4.
vmf.params: List. List of the parameters ot the fitted von-Mises distributions.
B: Integer value. Number of resampling.
cols.simul: Numerical value. Transformation matrix.
Y: Numerical vector or factor. Response.
func: Function. Variable selection function.
...: . Additionnal parameters passed to the func function.
Boost.coeffs: Numerical matrix. l2 normed matrix of predictors.
version: Character value. "lars" (no intercept value) or "glmnet" (first coefficient is the intercept value).

Value

Various types depending on the function.

Details

boost.normalize returns a numeric matrix whose colun are centered and l2 normalized.

boost.compcorrs returns a correlation like matrix computed using the corrfunc function.

boost.Xpass returns the transformation matrix.

boost.findgroups returns a list of groups or communities found using the group function.

boost.Xpass returns the transformation matrix.

boost.adjust returns the list of the parameters ot the fitted von-Mises distributions.

boost.random returns an array with the resampled datasets.

boost.apply returns a matrix with the coefficients estimated using the resampled datasets.

boost.select returns a vector with the proportion of times each variable was selected.

References

selectBoost: a general algorithm to enhance the performance of variable selection methods in correlated datasets, Frédéric Bertrand, Ismaïl Aouadi, Nicolas Jung, Raphael Carapito, Laurent Vallat, Seiamak Bahram, Myriam Maumy-Bertrand, Bioinformatics, 2020. doi:10.1093/bioinformatics/btaa855

Author

Frederic Bertrand, frederic.bertrand@utt.fr

Examples

set.seed(314)
xran=matrix(rnorm(200),20,10)
yran=rnorm(20)
xran_norm <- boost.normalize(xran)

xran_corr<- boost.compcorrs(xran_norm)

xran_corr_sign <- boost.correlation_sign(xran_corr)

xran_groups <- boost.findgroups(xran_corr, group=group_func_1, .3)
xran_groups_2 <- boost.findgroups(xran_corr, group=group_func_2, .3)

xran_Xpass <- boost.Xpass(nrow(xran_norm),ncol(xran_norm))

xran_adjust <- boost.adjust(xran_norm, xran_groups$groups, xran_corr_sign)

#Not meaningful, should be run with B>=100
xran_random <- boost.random(xran_norm, xran_Xpass, xran_adjust$vmf.params, B=5)

# \donttest{
xran_random <- boost.random(xran_norm, xran_Xpass, xran_adjust$vmf.params, B=100)
# }

xran_apply <- boost.apply(xran_norm, xran_random, yran, lasso_msgps_AICc)

xran_select <- boost.select(xran_apply)