Example of the update of XGBoost not working.

xgboost_failing_update.R

set.seed(12345)
library('data.table')
library('xgboost')

## Create simulation function
give_me_data <- function(nobs, delta = 0.05) {
  X_mat <- data.table(A = rnorm(nobs, 0, 1), B = rnorm(nobs, 0, 1))
  probs <- pmax(as.numeric(X_mat$A > 0) - 2*delta, 0) + delta
  X_mat <- as.matrix(X_mat)
  Y_vals <- rbinom(nobs, 1, probs)
  return(list(X_mat = X_mat, Y_vals = Y_vals))
}

nobs <- 20
niter <- 100
test_nobs <- 200

## Create a training set
data <- give_me_data(nobs)

## Create a test set
test_data <- give_me_data(test_nobs)

params <- list(objective = 'binary:logistic', nthread = 1)
dtrain <- xgb.DMatrix(data = data$X_mat, label = data$Y_vals)

previous_model <- xgb.train(
  data       = dtrain,
  params     = params,
  nrounds    = 200,
  xgb_model  = NULL
)

## Calculate result
result <- predict(previous_model, newdata = test_data$X_mat, type='response')

## Still works here, predictions look fine
print(result)

params <- modifyList(params, list(process_type = 'update', updater = 'refresh', refresh_leaf = TRUE))

## Update niter times
for (i in 1:niter) {
  data <- give_me_data(nobs)

  dtrain <- xgb.DMatrix(data = data$X_mat, label = data$Y_vals)
  previous_model <- xgb.train(
    data       = dtrain,
    params     = params,
    nrounds    = 200,
    xgb_model  = previous_model,
  )

  result <- predict(previous_model, newdata = test_data$X_mat, type='response')
  print(result)
  print(i)
}

## It works for the first iteration, then I get
## Error in xgb.train(data = dtrain, params = params, nrounds = 10, xgb_model = previous_model,: nrounds cannot be larger than 0 (nrounds of xgb_model)