GeDS (version 0.3.1)

crossv_GeDS: k-fold cross-validation

Description

crossv_GeDS performs k-fold cross-validation for tuning the relevant parameters of the NGeDS, GGeDS, NGeDSgam, and NGeDSboost models.

Value

Two data frames, best_params and results. best_params contains the best combination of parameters according to the cross-validated MSE. results presents the cross-validated MSE and the average number of internal knots across the folds for each possible combination of parameters, given the input parameters. In the case of model_fun = NGeDSboost, it also provides the cross-validated number of boosting iterations.

Arguments

formula

a description of the structure of the model structure, including the dependent and independent variables.

data

a data frame containing the variables referenced in the formula.

model_fun

the GeDS model to be fitted, that is, NGeDS, GGeDS, NGeDSgam or NGeDSboost.

parameters

to tune via cross-validation. These are: beta, phi and q in the case of NGeDS, GGeDS and NGeDSgam. In addition, for NGeDSboost, int.knots_init and shrinkage can also be tuned. Default values are int.knots_init_grid = c(0, 1, 2), shrinkage_grid = c(0.1, 0.5, 1), beta_grid = c(0.3, 0.5, 0.7), phi_grid = c(0.9, 0.95, 0.99), q_grid = c(2, 3)).

Examples

###################################################
# Generate a data sample for the response variable
# Y and the single covariate X
set.seed(123)
N <- 500
f_1 <- function(x) (10*x/(1+100*x^2))*4+4
X <- sort(runif(N, min = -2, max = 2))
# Specify a model for the mean of Y to include only a component
# non-linear in X, defined by the function f_1
means <- f_1(X)
# Add (Normal) noise to the mean of Y
Y <- rnorm(N, means, sd = 0.1)
data = data.frame(X = X, Y = Y)

if (FALSE) {
## NGeDS
# Define different combinations of parameters to cross-validate
param = list(beta_grid = c(0.5),
             phi_grid = c(0.9, 0.95),
             q_grid = c(2))

cv_NGeDS <- crossv_GeDS(Y ~ f(X), data = data, NGeDS, n = 3L,
                        parameters = param)

print(cv_NGeDS$best_params)
View(cv_NGeDS$results)

## NGeDSboost
param = list(int.knots_init_grid = c(1, 2),
             shrinkage_grid = 1,
             beta_grid = c(0.3, 0.5),
             phi_grid = c(0.95, 0.99),
             q_grid = 2)

cv_NGeDSboost <- crossv_GeDS(Y ~ f(X), data = data, NGeDSboost, n = 2L,
                             n_folds = 2L, parameters = param)

print(cv_NGeDSboost$best_params)
View(cv_NGeDSboost$results)
}