A novel design, but a little tedious?

[kangning-97]

Thanks for your perfect design. I'm verry sorry for bring some questions. Just as follows, is my code according with your design?

define original X and Y

Y <- df_pm25$PM2.5
X <- df_pm25 %>% select(-PM2.5,-DID,-sitename)
base_model <- SuperLearner(Y = Y, X = X,
SL.library = c("SL.glm","SL.glmnet"),# The first cluster models.
cvControl = list(V=2))

update new X according to the first cluster model

X_1 <- cbind(base_model$Z,X)
names(X_1)[1:2] <- paste0("X",1:2)
meta_model <- SuperLearner(Y = Y, X = X_1,
SL.library = c("SL.ranger","SL.glm"),# The second cluster models.
cvControl = list(V=2))

final

X_2 <- meta_model$Z %>% as.data.frame()
final_model <- SuperLearner(Y = Y, X = X_2,
SL.library = c("SL.nnls"),# The final cluster models.
cvControl = list(V=2))
pre <-final_model$SL.predict # prediction