CODEWITHSUNDEEP
rsqliter-caretxgboost
Swaroop Krothapalli
Swaroop Krothapalli

Reputation: 11

save xgb model trained using caret to database

I trained a xgb model using the caret package with traincontrol. I serialized and converted to character before saving the model to local SQLite database. I was able to save it. But while retrieving it from database and unserializing it, it throws an error unserialize(charToRaw(xgbModel))

Error in unserialize(charToRaw(selected_model$Model)) : ReadItem: unknown type 57, perhaps written by later version of R

Upvotes: 0

Views: 1135

Answers (1)

missuse
missuse

Reputation: 19716

To save the xgboost model trained by caret you need to save to raw. Here is an example:

library(mlbench) #for the data set
library(caret)
library(xgboost) 

data(Sonar)  

caret_model <- train(x = Sonar[,1:60],
                     y = Sonar$Class,
                     method = "xgbTree",
                     trControl = trainControl(method = "cv", number = 2),
                     tuneLength = 2)

save the raw model:

raw_model <- xgb.save.raw(caret_model$finalModel)

serialize that and save it to the local db. When you unserialise:

caret_model2 <- xgb.load(raw_model)

preds <- predict(caret_model2, as.matrix(Sonar[,1:60]))

head(preds)
#output
0.08729284 0.04738589 0.05628103 0.04275921 0.02574497 0.00655277

then you can use caret_model2 to predict.

This is mentioned here under "save and load models".

Upvotes: 0

Related Questions