crossvalidate.bigKRLS {bigKRLS} | R Documentation |
crossvalidate.bigKRLS
crossvalidate.bigKRLS(y, X, seed, Kfolds = NULL, ptesting = NULL, estimates_subfolder = NULL, ...)
y |
A vector of numeric observations on the dependent variable. Missing values not allowed. |
X |
A matrix of numeric observations of the independent variables. Factors, missing values, and constant vectors not allowed. |
seed |
Randomization seed to be used when partitioning data. |
Kfolds |
Number of folds for cross validation. Requires ptesting == NULL. Note KRLS assumes variation in each column; rare events or rarely observed factor levels may violate this assumption if Kfolds is too large given the data. |
ptesting |
Percentage of data to be used for testing (e.g., ptesting = 20 means 80% training, 20% testing). Requires Kfolds == NULL. Note KRLS assumes variation in each column; rare events or rarely observed factor levels may violate this assumptions if ptesting is too small given the data. |
estimates_subfolder |
If non-null, saves all model estimates in current working directory. |
... |
Additional arguments to be passed to bigKRLS() or predict(). E.g., crossvalidate.bigKRLS(y, X, derivative = FALSE) will run faster but compute fewer test stats comparing in and out of sample performance (because the marginal effects will not be estimated). |
bigKRLS_CV (list) Object of estimates and summary stats; summary() is defined. For train/test, contains a bigKRLS regression object and a predict object. For Kfolds, contains a nested series of training and testing models.
# y <- as.matrix(ChickWeight$weight) # X <- matrix(cbind(ChickWeight$Time, ChickWeight$Diet == 1), ncol = 2) # cv.out <- crossvalidate.bigKRLS(y, X, seed = 123, ptesting = 20) # cv.out$pseudoR2_oos # cv <- summary(cv.out) # cv$training.ttests # kcv.out <- crossvalidate.bigKRLS(y, X, seed = 123, Kfolds = 3) # kcv <- summary(kcv.out, digits = 3) # kcv$overview # kcv$training2.ttests # save.bigKRLS(kcv.out, "myKfolds") # load.bigKRLS("/path/to/myKfolds")