[R-package] current_iter in LightGBM R Model Objects Shares Memory Across Independent Models

[TaaviU]

Description

In the LightGBM R package, the current_iter attribute of independently trained models appears to share the same memory reference. This results in unexpected behavior where querying current_iter from one model affects the value of current_iter in another independent model.

Reproducible example

library(lightgbm)

m1 <- lgb.train(params = list(objective = "regression", metric = "rmse"),
                data = lgb.Dataset(matrix(runif(1000), ncol = 10), label = runif(100)),
                nrounds = 50)
n1 <- m1$current_iter() 

print(n1) # 50 as expected

m2 <- lgb.train(params = list(objective = "regression", metric = "rmse"),
                data = lgb.Dataset(matrix(runif(1000), ncol = 10), label = runif(100)),
                nrounds = 30)
n2 <- m2$current_iter()

print(n2) # 30 as expected

print(n1) # Expected: 50, Actual: 30

Environment info

LightGBM version: 4.5.0