使用R语言neuralnet包compute 函数问题

R版本3.52
使用neuralnet包进行神经网络的训练，完成后时用compute函数进行预测时出现问题
报错
Error in cbind(1, pred) %*% weights[[num_hidden_layers + 1]] :
  requires numeric/complex matrix/vector arguments


源程序

concrete<-read.csv("data.csv",header =  F,fileEncoding = "UTF-8",sep=",")
traindata<-concrete[1:50000,]
testdata<-concrete[50001:175814,]


traindata<-as.matrix(traindata)
testdata<-as.matrix(testdata)

library(neuralnet)
net <- neuralnet(x8~x1+x2+x3+x4+x5+x6+x7,traindata, hidden=c(5,3), threshold=0.005, learningrate = 0.1, algorithm = "rprop+", err.fct = "sse", act.fct = "logistic")



compute(net,testdata)