adaptive.weights function

Compute weights for use with adaptive lasso.

Fast computation of weights needed for adaptive lasso based on Gaussian family data.

adaptive.weights(x, y, nu = 1, weight.method = c("multivariate", "univariate"))

Arguments

• x: input matrix, of dimension nobs x nvars; each row is an observation vector.
• y: response variable.
• nu: non-negative tuning parameter
• weight.method: Should the weights be computed for multivariate regression model (only possible when the number of observations is larger than the number of parameters) or by individual marginal/"univariate" regression coefficients.

Returns

Returns a list with two elements: - weights: the computed weights - nu: the value of nu used for the computations

Details

The weights returned are 1/abs(beta_hat)^nu where the beta-parameters are estimated from the corresponding linear model (either multivariate or univariate).

Examples

set.seed(1)
x <- matrix(rnorm(50000), nrow=50)
y <- rnorm(50, mean=x[,1])
weights <- adaptive.weights(x, y)

if (requireNamespace("glmnet", quietly = TRUE)) {
    res <- glmnet::glmnet(x, y, penalty.factor=weights$weights)
    head(res)
}

References

Xou, H (2006). The Adaptive Lasso and Its Oracle Properties. JASA, Vol. 101.

See Also

glmnet

Author(s)

Claus Ekstrom claus@rprimer.dk