sig_soft: Variance estimators from Choi et al (2014).

In dcgerard/hose: Higher-order spectral estimators

Description

Returns a degrees of freedom corrected residual mean-squared error of a soft-thresholding estimator where the tuning parameter is chosen by a cross-validation-ish procedure.

Usage

sig_soft(Y, c_val = 2/3, k = 10, lambda_grid = NULL,
  print_update = FALSE)

Arguments

Y	The data matrix.
c_val	The ad-hoc adjustment to the degrees of freedom. Choi et al (2014) found that 2/3 worked well in simulations.
k	A positive integer. The fold for the soft-impute cross validation. Default is 10.
lambda_grid	A vector of positive numerics. The values of lambda to compute. The default is 20 values from the minimum to the maximum singular value of Y.
print_update	A logical. Should we print to the screen the status of the cross-validation-ish procedure at each iteration (TRUE) or not (FALSE)?

Details

See Choi et al (2014) for details. This seems to be the best estimation procedure so far, but also takes the longest.

You can try out different values of c_val with the outputs of sse and dfLambda

Value

sig2_est A positive numeric. The estimate of the variance.

sse A positive numeric. The sum of squared errors for estimated Y.

dfLambda A positive integer. The estimated df.

Author(s)

David Gerard

References

Choi, Yunjin, Jonathan Taylor, and Robert Tibshirani. "Selecting the number of principal components: Estimation of the true rank of a noisy matrix." arXiv preprint arXiv:1410.8260 (2014).

See Also

soft_cv for the cross-validation-ish procedure.

dcgerard/hose documentation built on Aug. 1, 2019, 12:11 a.m.