sparsereg3D.ncv: Model Evaluation with Nested Cross-Validation
In pejovic/sparsereg3D: An Implementation of Lasso Regression for 3D Soil Mapping.

Model Evaluation with Nested Cross-Validation

1 2	sparsereg3D.ncv(sparse.reg, lambda, step = FALSE, ols = FALSE, all = FALSE, lambda.1se = FALSE, w = NULL, alpha = 1)

`sparse.reg`	output from `pre.sparsereg3D` function
`lambda`	vector of regularization parameter values for lasso regression
`step`	logical. If TRUE stepwise procedure will be used when fitting OLS model
`ols`	logical. If TRUE model will be fitted with OLS insted of using lasso
`all`	logical. If TRUE a detailed output will be prepared.
`lambda.1se`	logical. If TRUE one sigma lambda rule will be used (largest lambda value with cv.err less than or equal to min(cv.err)+ SE).
`w`	weighted parameter (positive number) that controls the level of using observations from deeper layers as less informative. General weighted model is $w=1/(1+w*depth)$. This option is still under development, so it is not included in the function for model selection.
`alpha`	ElasticNet parameter. see glmnet function in glmnet package. Default is alpha = 1 indicating that the lasso penalty is used.
`seed`	random number generator

List of objects including:

RMSE: Root Mean Squared Error
R squared: Coefficient of determination
Accuracy (if all = TRUE). Accuracy measures computed for each step in outer cross-valudation loop within nested cross-validation.
data.prediction (if all = TRUE) Data frame containing design matrix extended with predictions obtain through nested cross-validation.
models (if all = TRUE) list of length nfolds containing models that corresponds to each step in outer cross-valudation loop within nested cross-validation.