Require finite/non-NA/non-NaN training/testing data

[martinju]

I believe all types of approaches (perhaps except approach=independence) requires data to be finite, non-NA and non-NaN. It therefore makes sense to check for this in the check_features function call. Maybe after this code?:

shapr/R/preprocess_data.R

Lines 260 to 269 in c7455a1

	# Check if the features all have class "integer", "numeric" or "factor
	if (!all(f_list_1$classes %in% c("integer", "numeric", "factor"))) {
	invalid_class <- which(!(f_list_1$classes %in% c("integer", "numeric", "factor")))
	stop(paste0(
	"Feature(s) ", paste0(f_list_1$labels[invalid_class], collapse = ", "), " in ", name_1, " and ", name_2,
	" is not of class integer, numeric or factor."
	))
	}