Skip to contents

Initial coefficients

Source: R/group_lasso.R
estim_initial_coefs.Rd

Estimate initial coefficients.

Usage

.estim_initial_coefs(
  x,
  y,
  family = "gaussian",
  alpha_init = 0,
  group = NULL,
  foldid = NULL,
  nfolds = 10L,
  lambda = NULL,
  silent = FALSE
)

Arguments

x

\(n_0 \times p\) predictor matrix, containing only numerical values (continuous, integer, or binary), where \(n_0\) is the number of observations used for model training and \(p\) is the number of predictors

y

response vector of length \(n_0\), containing numerical values (family="gaussian"), integer values (family="poisson"), binary values (family="binomial"), or a survival object created with survival::Surv() (family="cox"), where \(n_0\) is the number of observations used for model training

family

character string "gaussian", "binomial", "poisson", or "cox"

alpha_init

A scalar specifying the method used for obtaining initial coefficients:

  • a numeric scalar in the unit interval (\(0 \leq\) alpha_init \(\leq 1\)) to define the mixing parameter for elastic net regression (default: ridge penalisation with alpha_init=0);

  • the character scalar "pearson", "spearman", or "kendall" to use initial correlation coefficients (not implemented for family="cox")

  • the character scalar"multiridge" to use multi-penalty ridge regression with one penalty for each group (not implemented for family="poisson" or overlapping groups),

  • NA to set all initial coefficients equal to 1

group

\(p\)-dimensional integer vector with entries in \(\{1, \ldots, q\}\)

foldid

\(n_0\)-dimensional vector containing the fold identifiers (minimum \(1\), maximum nfolds)

nfolds

positive integer specifying the number of folds (minimum \(3\), maximum \(n\))

NB: If foldid is provided, nfolds is overwritten by max(foldid).

lambda

numeric scalar, or NULL (determined by cross-validation)

silent

Should messages from glmnet::glmnet() and glmnet::cv.glmnet() be suppressed? (logical scalar, FALSE or TRUE)

Value

Returns a list with two slots:

  • coef: numeric vector of length \(p\) (estimated coefficients)

  • lambda: non-negative numeric scalar (optimised regularisation parameter) or NULL

Details

This function is called by corila(). It calls glmnet::cv.glmnet() or glmnet::glmnet() for an initial lasso, ridge, or elastic net regression, multiridge() for an initial multi-penalty ridge regression, or stats::cor() for initial correlation coefficients.

Examples

# simulate data
set.seed(1)
n <- 50L
p <- 10L
x <- matrix(rnorm(n * p), nrow = n, ncol = p)
beta <- rbinom(n = p, size = 1L, prob = 0.5) * rnorm(p)
y <- drop(x %*% beta)

# initial correlation coefficients
.estim_initial_coefs(x = x,
                     y = y,
                     family = "gaussian",
                     alpha_init = "spearman",
                     group = NULL,
                     foldid = NULL,
                     nfolds = 10L,
                     lambda = NULL)
#> $coef
#>  [1]  0.53478992 -0.50991597  0.31870348  0.19558223 -0.09483794 -0.02713085
#>  [7] -0.18953181 -0.42223289 -0.03990396 -0.30554622
#> 
#> $lambda
#> NULL
#> 

# initial regression coefficients (cross-validating lambda)
foldid <- sample(seq_len(10L), size = n, replace = TRUE)
.estim_initial_coefs(x = x,
                     y = y,
                     family = "gaussian",
                     alpha_init = 0.0,
                     group = NULL,
                     foldid = foldid,
                     nfolds = 10L,
                     lambda = NULL,
                     silent = TRUE)
#> $coef
#>  [1]  1.47303510 -1.30312590  0.04308875  1.05078015 -0.02502813 -0.02995143
#>  [7] -0.85606093 -0.92824916  0.04838767 -0.39441178
#> 
#> $lambda
#> [1] 0.126489
#> 

# initial regression coefficients (using fixed lambda)
.estim_initial_coefs(x = x,
                     y = y,
                     family = "gaussian",
                     alpha_init = 0.0,
                     group = NULL,
                     foldid = NULL,
                     nfolds = 10L,
                     lambda = 0.2)
#> $coef
#>  [1]  1.41315853 -1.26524541  0.06226520  0.99767196 -0.03630484 -0.04342165
#>  [7] -0.81640523 -0.90602485  0.04799027 -0.38908914
#> 
#> $lambda
#> [1] 0.2
#>