r - 在插入符号的训练函数中并行调整 xgboost 模型时出错

Question

我正在尝试在插入符号中为 xgboost 模型运行一些交叉验证调整。我有一个很大的调整网格，所以我想并行运行它。我将数据设置为稀疏矩阵，设置调整网格，并行处理，然后尝试运行train，但每次都收到连接错误。如果我禁用并行选项，它将运行得很好。这不是我的数据，因为这个样本数据和我的实际数据都有同样的问题。可能是什么原因造成的？我也很好奇为什么我需要在train函数中定义 y 标签，而 xgb.Dmatrixdtrain已经有一个指向它的指针。

示例数据位于 DALEX 包中，即公寓数据集。

library(caret)
library(xgboost)
library(Matrix)
library(DALEX) # get access to the sample data called "apartments"
library(doParallel)

x.train <- sparse.model.matrix(m2.price ~. -1 , data = apartments)
dtrain <- xgb.DMatrix(x.train, label = apartments$m2.price)

grid = expand.grid(
  nrounds = 500,
  eta = seq(.002,0.004,by = .002),
  max_depth = seq(2, 4, by = 2),
  gamma = 0, 
  colsample_bytree = 1,
  min_child_weight = seq(8, 10, by = 2),
  subsample = 0.5
)

# set cross validation
fitControl = trainControl(
  method = "cv",
  number = 5
)

# set up parallel processing
cl <- makeCluster(detectCores())
registerDoParallel(cl)
getDoParWorkers()


Tune = train(x = dtrain, y = apartments$m2.price,
             trControl = fitControl,
             tuneGrid = grid,
             method = "xgbTree",
             na.action = na.pass
)

Error in serialize(data, node$con) : error writing to connection

标签： rr-caretxgboostdoparallel