> # read the datafile
> toyota.df <- read.csv("ToyotaCorolla.csv")
>
> # set variable names
> vars = c("Age_08_04", "KM", "Fuel_Type", "Automatic", "Powered_Windows")
>
> # normalize data
> toyota.df.norm <- preProcess(toyota.df[, c(vars, "Price")], method=c("range"))
> toyota.df[, c(vars, "Price")] <- predict(toyota.df.norm, toyota.df[, c(vars, "Price")])
>
> toyota.df.norm <- preProcess(toyota.df[, c(vars, "Age_08_04")], method=c("range"))
> toyota.df.norm <- preProcess(toyota.df[, c(vars, "KM")], method=c("range"))
>
> # partition the data 60% training, 40% validating
> set.seed(2)
> training = sample(row.names(toyota.df), dim(toyota.df)[1]*0.6)
> validation = setdiff(row.names(toyota.df), training)
>
> # when y has multiple classes - need to dummify
> trainData <- cbind(toyota.df[training,c(vars)],
+                    class.ind(toyota.df[training,]$Fuel_Type),
+                    class.ind(toyota.df[training,]$Automatic),
+                    class.ind(toyota.df[training,]$Powered_Windows),
+                    toyota.df[training,]$Price)
> names(trainData) <- c(vars, paste("Fuel_Type_", c("CNG", "Diesel", "Petrol"), sep = ""),
+                       paste("Automatic_", c("No", "Yes"), sep = ""),
+                       paste("PowerWindow_", c("No", "Yes"), sep = ""),
+                       "Price")
>
> validData <- cbind(toyota.df[validation,c(vars)],
+                    class.ind(toyota.df[validation,]$Fuel_Type),
+                    class.ind(toyota.df[validation,]$Automatic),
+                    class.ind(toyota.df[validation,]$Powered_Windows),
+                    toyota.df[validation,]$Price)
> names(validData) <- c(vars, paste("Fuel_Type_", c("CNG", "Diesel", "Petrol"), sep = ""),
+                       paste("Automatic_", c("No", "Yes"), sep = ""),
+                       paste("PowerWindow_", c("No", "Yes"), sep = ""),
+                       "Price")
>
> # run nn with 2 hidden nodes
> # use hidden= with a vector of integers specifying number of hidden nodes in each layer
> nn <- neuralnet(Price ~ Age_08_04 + KM + Fuel_Type_CNG + Fuel_Type_Diesel + Fuel_Type_Petrol
+                 + Automatic_No + Automatic_Yes + PowerWindow_No + PowerWindow_Yes,
+                 data = trainData, hidden = 2)
>
> # draw nn diagram
> plot(nn, rep="best")

> # predict the values
> training.prediction=neuralnet::compute(nn, trainData[,-c(10)])
Error in neurons[[i]] %*% weights[[i]] :
  requires numeric/complex matrix/vector arguments
> validation.prediction=neuralnet::compute(nn, validData[,-c(10)])
Error in neurons[[i]] %*% weights[[i]] :
  requires numeric/complex matrix/vector arguments