validation_data=NULL

[ablanda]

Hi, I try to set validation_data=NULL, but after 1 Layer it stops by this message: Error in alloc.col(dt1, length(colnames(dt1)) + length(cols)) :
alloccol has been passed a NULL dt
Could you help me? Thanks

[Laurae2]

Do you have a reproducible example?

[ablanda]

library(Laurae)
library(Matrix)
library(xgboost)
library(data.table)

data("airquality")
airquality<-na.omit(airquality)
nrow(airquality)

label_train<-airquality[,1]
train<-as.data.table(airquality[,-1])

m<-CascadeForest(
training_data = train,
validation_data = NULL,
training_labels = label_train,
folds = folds,
boosting = FALSE,
nthread = 3,

cascade_lr = 1,
training_start = NULL,
validation_start = NULL,
cascade_forests = c(rep(4, 2), 0),
cascade_trees = 100,
cascade_rf = 4,
objective = "reg:linear",
eval_metric = Laurae::df_rmse,
multi_class = FALSE,
early_stopping = 2,
maximize = TRUE,
verbose = TRUE,
essentials = TRUE
)

I think you sholud change at row 306 of Cascade forest's script in validation_data <- if (!is.null(validation_data)) {
Laurae::DTcbind(validation_data, model[[i]]$valid_preds)}
else { NULL }

[Laurae2]

@ablanda My home repo is currently out of sync with the main repo, so I will fix it later.

You can temporarily fix this by doing this like you said:

https://github.com/Laurae2/Laurae/blob/master/R/CascadeForest.R#L306-L307

if (!is.null(validation_data)) {
    validation_data <- Laurae::DTcbind(validation_data, model[[i]]$valid_preds) # Routine overwrites if column names are identical
    validation_start <- if (boosting) {rowMeans(model[[i]]$valid_preds)} else {NULL}
}

do the same for the if here: https://github.com/Laurae2/Laurae/blob/master/R/CascadeForest.R#L445-L446