1

I have the following model to predict the price of houses in a particular neighborhood :

set.seed(100)
index_1<-sample(1:nrow(data),round(nrow(data)*0.9))
train<-data[index_1,] #578 obs.
test<-data[-index_1,] #62 obs.

NModel <- neuralnet(price ~  x1 + x2 + x3 + x4, data=train_group, hidden=c(5), linear.output=FALSE, threshold =0.01, rep=20, learningrate = 0.25 )

However I have been changing my learning rate from 0.25 to 1 and nothing has changed on my RMSE. It neither gets worse nor gets better, it stays exactly the same, even when changing a learning rate. Does anyone have any hints of what may be happening?

desertnaut
  • 57,590
  • 26
  • 140
  • 166
user14738548
  • 167
  • 1
  • 1
  • 7

1 Answers1

0

neuralnet function | R Documentation

learningrate is used only for traditional backpropagation. Try adding the argument algorithm = 'backprop' and then see what different values for learningrate do.

Jacob Myer
  • 479
  • 5
  • 22
  • Thank you very much, Jacob! Can I apply the backpropagation method for this type of neuronal network, in particular, without a problem? Or does the backpropagation method apply only to some types of neural network? – user14738548 Dec 03 '20 at 18:35
  • You should be able to. Backpropagation is just one algorithm that can be used to compute the node weights of the neural network. It might be worth trying some other algorithms (without the `learningrate` argument) and comparing the results of the model. – Jacob Myer Dec 03 '20 at 18:45