CODEWITHSUNDEEP
rmachine-learning
Alexandre Halm
Alexandre Halm

Reputation: 989

R: Reduce size of rpart object

I am trying to aggregate in a list multiple CART models built with rpart::rpart .

I just realised that each model stores quite a lot of (meta?)data in $terms and $where (over 10MB of data per model in my case) which makes my final list un-manageable.

It seems to me that the summary given by print(my_rpart_object) should be sufficient to describe the object and run predictions, so I wonder if there is a way to trim / compress rpart trees?

Upvotes: 2

Views: 456

Answers (2)

CrossValidatedTrading
CrossValidatedTrading

Reputation: 113

I was struggling with this as well. I found setting the "where" element of the rpart tree to NULL significantly reduced the tree's memory footprint:

rpart_model <- rpart(...)
rpart_model$where <- NULL

Upvotes: 1

Alexandre Halm
Alexandre Halm

Reputation: 989

Found it: each rpart object was carrying an environment with it. To remove it :

rpart_model <- rpart(...)
environment(rpart_model$terms) <- NULL

List of 21 part objects went from 1.2GB to 8MB.

Upvotes: 2

Related Questions