1

I am running the following command on R:

tg <- expand.grid(layer1 = 22, layer2 =0, layer3 =0)
tc <- trainControl(method = "repeatedcv" , number = 7, repeats = 5)
registerDoMC(cores = 30)
Model <- train( x~ ., data, method ="neuralnet", trControl = tc, tuneGrid = tg)

I am running the above on 32 cores, 64 Gb RAM on amazon instance. The issue is that code runs for 2 hours with no output. When i change method to "nnet" it runs normally( i know that neuralnet requires more processing)

But i would stay for 2 hours with no outputs for "neuralnet" and "qrnn" Is this normal? Or there is something i do not know?

Omar Elrafei
  • 11
  • 1
  • Could you try h2o and provide a reproducible case w/ input data size ,network architecture? I'd like to go deeper for performance optimization. – Patric Dec 13 '15 at 10:45
  • i tried h20, performance is so much faster. but still, is there any reason why "neuralnet" and "qrnn" take forever? i am training 1017209 rows with 14 predictors (response variable is included) and i choose hidden layers to be 26 (best rmse i could get.) – Omar Elrafei Dec 14 '15 at 19:48
  • I will try your case tomorrow;) could you make a reproducible case w/ input data? – Patric Dec 14 '15 at 19:52
  • @Patric sorry for my ignorance, what do you mean with reproducible case w/input data? – Omar Elrafei Dec 14 '15 at 20:06
  • See this link http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example and [ask] – Minnow Dec 15 '15 at 00:17
  • Did you try to cut down the testing data or even run it in parallel? – Ka_Papa Nov 20 '17 at 02:50

0 Answers0