Perform (repeated) K-fold cross-validation for pense().

adapense_cv() is a convenience wrapper to compute adaptive PENSE estimates.

pense_cv(
x,
y,
standardize = TRUE,
lambda,
cv_k,
cv_repl = 1,
cv_metric = c("tau_size", "mape", "rmspe", "auroc"),
fit_all = TRUE,
cl = NULL,
...
)

adapense_cv(x, y, alpha, alpha_preliminary = 0, exponent = 1, ...)

## Arguments

x n by p matrix of numeric predictors. vector of response values of length n. For binary classification, y should be a factor with 2 levels. whether to standardize the x variables prior to fitting the PENSE estimates. Can also be set to "cv_only", in which case the input data is not standardized, but the training data in the CV folds is scaled to match the scaling of the input data. Coefficients are always returned on the original scale. This can fail for variables with a large proportion of a single value (e.g., zero-inflated data). In this case, either compute with standardize = FALSE or standardize the data manually. optional user-supplied sequence of penalization levels. If given and not NULL, nlambda and lambda_min_ratio are ignored. number of folds per cross-validation. number of cross-validation replications. either a string specifying the performance metric to use, or a function to evaluate prediction errors in a single CV replication. If a function, the number of arguments define the data the function receives. If the function takes a single argument, it is called with a single numeric vector of prediction errors. If the function takes two or more arguments, it is called with the predicted values as first argument and the true values as second argument. The function must always return a single numeric value quantifying the prediction performance. The order of the given values corresponds to the order in the input data. If TRUE, fit the model for all penalization levels. Otherwise, only at penalization level with smallest average CV performance. a parallel cluster. Can only be used if ncores = 1, because multi-threading can not be used in parallel R sessions on the same host. Arguments passed on to pense nlambdanumber of penalization levels. lambda_min_ratioSmallest value of the penalization level as a fraction of the largest level (i.e., the smallest value for which all coefficients are zero). The default depends on the sample size relative to the number of variables and alpha. If more observations than variables are available, the default is 1e-3 * alpha, otherwise 1e-2 * alpha. nlambda_enpynumber of penalization levels where the EN-PY initial estimate is computed. penalty_loadingsa vector of positive penalty loadings (a.k.a. weights) for different penalization of each coefficient. Only allowed for alpha > 0. enpy_lambdaoptional user-supplied sequence of penalization levels at which EN-PY initial estimates are computed. If given and not NULL, nlambda_enpy is ignored. other_startsa list of other staring points, created by starting_point(). If the output of enpy_initial_estimates() is given, the starting points will be shared among all penalization levels. Note that if a the starting point is specific to a penalization level, this penalization level is added to the grid of penalization levels (either the manually specified grid in lambda or the automatically generated grid of size nlambda). If standardize = TRUE, the starting points are also scaled. interceptinclude an intercept in the model. bdpdesired breakdown point of the estimator, between 0 and 0.5. cctuning constant for the S-estimator. Default is to chosen based on the breakdown point bdp. Does not affect the estimated coefficients, only the estimated scale of the residuals. epsnumerical tolerance. explore_solutionsnumber of solutions to compute up to the desired precision eps. explore_tolnumerical tolerance and maximum number of iterations for exploring possible solutions. The tolerance should be (much) looser than eps to be useful, and the number of iterations should also be much smaller than the maximum number of iterations given via algorithm_opts. explore_itnumerical tolerance and maximum number of iterations for exploring possible solutions. The tolerance should be (much) looser than eps to be useful, and the number of iterations should also be much smaller than the maximum number of iterations given via algorithm_opts. max_solutionsonly retain up to max_solutions unique solutions per penalization level. comparison_tolnumeric tolerance to determine if two solutions are equal. The comparison is first done on the absolute difference in the value of the objective function at the solution If this is less than comparison_tol, two solutions are deemed equal if the squared difference of the intercepts is less than comparison_tol and the squared $$L_2$$ norm of the difference vector is less than comparison_tol. add_zero_basedalso consider the 0-based regularization path. See details for a description. enpy_specificuse the EN-PY initial estimates only at the penalization level they are computed for. See details for a description. sparseuse sparse coefficient vectors. ncoresnumber of CPU cores to use in parallel. By default, only one CPU core is used. May not be supported on your platform, in which case a warning is given. algorithm_optsoptions for the MM algorithm to compute the estimates. See mm_algorithm_options() for details. mscale_optsoptions for the M-scale estimation. See mscale_algorithm_options() for details. enpy_optsoptions for the ENPY initial estimates, created with the enpy_options() function. See enpy_initial_estimates() for details. cv_objectivedeprecated and ignored. See pense_cv() for estimating prediction performance via cross-validation. elastic net penalty mixing parameter with $$0 \le \alpha \le 1$$. alpha = 1 is the LASSO penalty, and alpha = 0 the Ridge penalty. alpha parameter for the preliminary estimate. the exponent for computing the penalty loadings based on the preliminary estimate.

## Value

a list with components:

lambda

the sequence of penalization levels.

cvres

data frame of average cross-validated performance.

cv_replications

matrix of cross-validated performance metrics, one column per replication. Rows are in the same order as in cvres.

call

the original call.

estimates

the estimates fitted on the full data. Same format as returned by pense().

the object returned by adapense_cv() has additional components

preliminary

the CV results for the preliminary estimate.

penalty_loadings

## Details

The built-in CV metrics are

"tau_size"

$$\tau$$-size of the prediction error, computed by tau_size() (default).

"mape"

Median absolute prediction error.

"rmspe"

Root mean squared prediction error.

"auroc"

Area under the receiver operator characteristic curve (actually 1 - AUROC). Only sensible for binary responses.

adapense_cv() is a convenience wrapper which performs 3 steps:

1. compute preliminary estimates via pense_cv(..., alpha = alpha_preliminary),

2. computes the penalty loadings from the estimate beta with best prediction performance by adapense_loadings = 1 / abs(beta)^exponent, and

3. compute the adaptive PENSE estimates via pense_cv(..., penalty_loadings = adapense_loadings).

pense() for computing regularized S-estimates without cross-validation.

coef.pense_cvfit() for extracting coefficient estimates.

plot.pense_cvfit() for plotting the CV performance or the regularization path.

Other functions to compute robust estimates with CV: pensem_cv(), regmest_cv()

Other functions to compute robust estimates with CV: pensem_cv(), regmest_cv()

## Examples

# Compute the adaptive PENSE regularization path for Freeny's
# revenue data (see ?freeny)
data(freeny)
x <- as.matrix(freeny[ , 2:5])

## Either use the convenience function directly ...
ada_convenience <- adapense_cv(x, freeny$y, alpha = 0.5, cv_repl = 2, cv_k = 4) ## ... or compute the steps manually: # Step 1: Compute preliminary estimates with CV preliminary_estimate <- pense_cv(x, freeny$y, alpha = 0,
cv_repl = 2, cv_k = 4)
plot(preliminary_estimate, se_mult = 1)

# Step 2: Use the coefficients with best prediction performance
ada_manual <- pense_cv(x, freeny$y, alpha = 0.5, cv_repl = 2, cv_k = 4, penalty_loadings = pen_loadings) # Visualize the prediction performance and coefficient path of # the adaptive PENSE estimates (manual vs. automatic) def.par <- par(no.readonly = TRUE) layout(matrix(1:4, ncol = 2, byrow = TRUE)) plot(ada_convenience$preliminary)