dailyevalgist.r

mape <- function(actual, predicted, f) {
  100 * mean(abs(f(actual) - f(predicted)) / f(actual))
}
 
mse <- function(actual, predicted, f) {
  mean((f(actual) - f(predicted)) ^ 2)
}
 
mae <- function(actual, predicted, f) {
  mean(abs(f(actual) - f(predicted)))
}

linear.model <- lm(Zt ~ (Z1 * Z2 * Z3)^3, trainer)
linear.preds <- predict(linear.model, tester)
linear.mae <- mae(tester$Zt, linear.preds, identity)
linear.mape <- mape(tester$Zt, linear.preds, identity)
linear.mse <- mse(tester$Zt, linear.preds, identity)


polykernelmodel <- ksvm(Zt ~ (Z1 * Z2 * Z3)^3, trainer, kernel='polydot')
polykernelpreds <- predict(polykernelmodel, tester)
polykernel.mae <- mae(tester$Zt, polykernelpreds, identity)
polykernel.mape <- mape(tester$Zt, polykernelpreds, identity)
polykernel.mape <- mape(tester$Zt, polykernelpreds, identity)

rbfkernelmodel <- ksvm(Zt ~ (Z1 * Z2 * Z3)^3, trainer)
rbfkernelpreds <- predict(rbfkernelmodel, tester)
rbfkernel.mae <- mae(tester$Zt, rbfkernelpreds, identity)
rbfkernel.mape <- mape(tester$Zt, rbfkernelpreds, identity)
rbfkernel.mape <- mape(tester$Zt, rbfkernelpreds, identity)

network <- neuralnet(Zt ~ Z1 + Z2 + Z3, training_normalized_data_set, hidden=3)
network.preds <- compute(network,
                             testing_normalized_data_set[,c('Z1','Z2','Z3')])
network.preds <- Zt_norms[[2]](network.preds$net.result)
network.mse <- mse(tester$Zt, network.preds, identity)
network.mae <- mae(tester$Zt, network.preds, identity)
network.mape <- mape(tester$Zt, network.preds, identity)