Neural Network in R

---

Neural Network Example

Loading data and libraries.

Read all the libraryies you need.

Build labels for classification

In iris dataset, there are three types of flowers, which means we’re now dealing with a multi-class classification.

xor <- model.matrix(~ Species, Data)
xor[,1] <- ifelse(xor[,2] %in% 0 & xor[,3] %in% 0, 1, 0)
colnames(xor)[1] <- 'Speciessetosa'
head(xor)

##   Speciessetosa Speciesversicolor Speciesvirginica
## 1             1                 0                0
## 2             1                 0                0
## 3             1                 0                0
## 4             1                 0                0
## 5             1                 0                0
## 6             1                 0                0

Data <- cbind(xor, Data[,-5])

Split data for training

set.seed(30)
DataFinal <- Data[sample(nrow(Data)),]

Size <- floor(0.7*nrow(Data))

TrainSet <- DataFinal[1:Size, ]
TestSet <- DataFinal[c(Size+1):nrow(Data), ]

Train NN model

We can us neuralnet() to train a NN model. Also, the train() function from caret can help us tune parameters. We can plot the result to see which set of parameters is fit our data the best.

FormulaNN <- Speciessetosa + Speciesversicolor + Speciesvirginica ~ Sepal.Length + Sepal.Width + Petal.Length + Petal.Width

Model <- train(form=FormulaNN,     
               data=TrainSet,           
               method="neuralnet",   
               ### Parameters for layers
               tuneGrid = expand.grid(.layer1=c(1:4), .layer2=c(0:4), .layer3=c(0)),               
               ### Parameters for optmization
               learningrate = 0.01,  
               threshold = 0.01,     
               stepmax = 50000         
)

## Warning in nominalTrainWorkflow(x = x, y = y, wts = weights, info =
## trainInfo, : There were missing values in resampled performance measures.

plot(Model)

Bulid best model

Build model based on the result from the plot above and plot the model.

BestModel <- neuralnet(formula = FormulaNN, 
                 data = TrainSet,
                 hidden = c(1,2),     
                 learningrate = 0.01, 
                 threshold = 0.01,   
                 stepmax = 50000       
                 
)

plot(BestModel, rep = 'best')

Measure the performance

Use compute() to calculate the prediction of the model.

pred <-  compute(BestModel, TestSet[,4:7])
pred <- round(pred$net.result)

Result <- ifelse(pred[,1] %in% 1, 'setosa',
                 ifelse(pred[,2] %in% 1, 'versicolor', 'virginica'))
Species <- ifelse(TestSet[,1] %in% 1, 'setosa',
                 ifelse(TestSet[,2] %in% 1, 'versicolor', 'virginica'))

table(Result, Species)

##             Species
## Result       setosa versicolor virginica
##   setosa         15          0         0
##   versicolor      0         16         0
##   virginica       0          0        14