Importance statistics based on the square-root lasso

Computes the signed maximum statistic $$W_j = \max(Z_j, \tilde{Z}_j) \cdot \mathrm{sgn}(Z_j - \tilde{Z}_j),$$ where \(Z_j\) and \(\tilde{Z}_j\) are the maximum values of \(\lambda\) at which the jth variable and its knockoff, respectively, enter the SQRT lasso model.

stat.sqrt_lasso(X, X_k, y, ...)

Arguments

X

n-by-p matrix of original variables.

X_k

n-by-p matrix of knockoff variables.

y

vector of length n, containing the response variables of numeric type.

...

additional arguments specific to slim.

Value

A vector of statistics \(W\) of length p.

Details

With default parameters, this function uses the package RPtests to run the SQRT lasso. By specifying the appropriate optional parameters, one can use different Lasso variants including Dantzig Selector, LAD Lasso, SQRT Lasso and Lq Lasso for estimating high dimensional sparse linear models.

For a complete list of the available additional arguments, see RPtests::sqrt_lasso().

Examples

set.seed(2024)
n=80; p=100; k=10; Ac = 1:k; Ic = (k+1):p
X = generate_X(n=n,p=p)
y <- generate_y(X, p_nn=k, a=3)
Xk = create.shrink_Gaussian(X = X, n_ko = 10)
res1 = knockoff.filter(X, y, Xk, statistic = stat.sqrt_lasso,
                       offset = 1, fdr = 0.1)
res1
#> Call:
#> knockoff.filter(X = X, y = y, Xk = Xk, statistic = stat.sqrt_lasso, 
#>     fdr = 0.1, offset = 1)
#> 
#> Selected variables:
#> [1]  1  2 10
#> 
#> Frequency of selected variables from 30 knockoff copys:
#>   [1] 10 10  0  0  0  0  0  0  0 10  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
#>  [26]  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
#>  [51]  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
#>  [76]  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
perf_eval(res1$shat,Ac,Ic)
#> [1] 0.3 0.0