neuralnet-square-root.R

if (!require(neuralnet)) {install.packages("neuralnet")}

library(neuralnet)

#nn for learning how to calculate a square root

set.seed(1)
train.x <- sample(1:100, 50, replace = T)
head(train.x)

train.y <- sqrt(train.x)
head(train.y)

train.xy <- as.data.frame(cbind(train.x, train.y))
head(train.xy)


net.sqrt <- neuralnet(train.y ~ train.x,
                    train.xy, 
                    hidden=10,
                    threshold = 0.01 #1% error threshold
                    )

plot(net.sqrt)

test.x <- as.data.frame((1:10)^2) #new set of squared numbers
test.y <- compute(net.sqrt, test.x) #run through ANN
test.y


##results table
net.table <- cbind(test.x,
                   sqrt(test.x),
                   round(test.y$net.result, 2))
colnames(net.table) <- c("X", "Y-Exp", "Y-ANN")
net.table