RAM usage of xgboost GPU in R

xgboost_GPU_RAM_usage.R

library(xgboost)

set.seed(1)
N <- 1000000
p <- 100
pp <- 25
X <- matrix(runif(N * p), ncol = p)
betas <- 2 * runif(pp) - 1
sel <- sort(sample(p, pp))
m <- X[, sel] %*% betas - 1 + rnorm(N)
y <- rbinom(N, 1, plogis(m))

format(object.size(X), units = "Mb")

trainer <- function(n_cpus, n_gpus, n_iterations, n_depth) {
  
  if (n_gpus == 0) {
    
    pt <- proc.time()
    set.seed(11111)
    model <- xgb.train(list(objective = "binary:logistic", nthread = n_cpus, eta = 0.10, max_depth = n_depth, max_bin = 64, tree_method = "hist"),
                       dtrain, nrounds = n_iterations, verbose = 0)
    my_time <- proc.time() - pt
    
  } else {
    
    pt <- proc.time()
    set.seed(11111)
    model <- xgb.train(list(objective = "gpu:binary:logistic", nthread = n_cpus, eta = 0.10, max_depth = n_depth, max_bin = 64, tree_method = "gpu_hist", n_gpus = n_gpus),
                       dtrain, nrounds = n_iterations, verbose = 0)
    my_time <- proc.time() - pt
    
  }
  
  return(my_time)
  
}

dtrain <- xgb.DMatrix(X, label = y)
wl <- list(test = dtest)
gpus_to_use_old <- 1

gpus_to_use <- 1
n_iters_to_use <- 5
depth_to_use <- 1:12
gc(verbose = FALSE)

if (gpus_to_use != gpus_to_use_old) {
  rm(dtrain)
  gc(verbose = FALSE)
  dtrain <- xgb.DMatrix(X, label = y)
  gc(verbose = FALSE)
}

gpus_to_use_old <- gpus_to_use

for (i in depth_to_use) {
  trainer(gpus_to_use, gpus_to_use, n_iters_to_use, i)
  cat("Record for Depth ", i, "!", sep = "")
  Sys.sleep(5)
  cat(" | Refreshed!\n", sep = "")
  gc(verbose = FALSE)
  Sys.sleep(3)
}