source('BuildEmulator/BuildEmulator.R')
simulator = function(x) {
  sin(2*x) * exp(-0.5 * x^2)
}
df_simulator = data.frame(x = seq(-3, 3, 0.1), y = simulator(seq(-3, 3, 0.1)))
# Normalize your input x to [-1, 1]
df_simulator$x = 2*(df_simulator$x - min(df_simulator$x))/(max(df_simulator$x) - min(df_simulator$x)) - 1
ggplot(df_simulator, aes(x=x, y=y)) + geom_line()
# Data frame with design set and noise
tData = data.frame(x = seq(-2.5, 2.5, length.out = 15), Noise = rnorm(15, 0, 0.4), y = simulator(seq(-2.5, 2.5, length.out = 15)))
# Normalize your input x to [-1, 1]
tData$x = 2*(tData$x - (-3))/(3 - (-3)) - 1
print(tData)
ggplot(tData, aes(x = x, y=y)) + geom_point()
cands <- names(tData)[1]
# Construct a mean function for GP emulator.
myem.lm <- EMULATE.lm(Response=names(tData)[3], tData=tData, tcands=cands,tcanfacs=NULL,TryFouriers=FALSE, maxOrder=2, maxdf = 2)
# Construct GP emulator.
myem.gp <- EMULATE.gpstan(meanResponse=myem.lm, nugget=0.0001, tData=tData, additionalVariables=cands, 
                          FastVersion=TRUE)
# Diagnostics: LOO
tLOOs <- LOO.plot(StanEmulator = myem.gp, ParamNames=names(tData)[1])
# Predictions for a new data set.
NewData <- data.frame(x=df_simulator$x)
head(NewData)
valid <- EMULATOR.gpstan(data.frame(x=df_simulator$x), Emulator = myem.gp)
fit.valid <- cbind(valid$Expectation, valid$Expectation - 2*sqrt(valid$Variance),
                   valid$Expectation + 2*sqrt(valid$Variance))
ValidPlot(fit.valid, matrix(df_simulator$x, ncol = 1), df_simulator$y, 
          interval = c(), axis = 1, heading = "", xrange = c(-1, 1), ParamNames = c('x'))