#table type should be matrix instead of data.frame for ForeCA calculations
#trainData <- as.matrix(read.csv("D:\\trainData.csv", header=TRUE, sep=";", na.strings=c(".", "NA", "", "?"), strip.white=TRUE, encoding="UTF-8", check.names=TRUE))
#trainDataFT <- as.matrix(read.csv("D:\\fronttestData.csv", header=TRUE, sep=";", na.strings=c(".", "NA", "", "?"), strip.white=TRUE, encoding="UTF-8", check.names=TRUE))
load("D:\\trainData.RData")

library(ForeCA)

RegressionToClasses01 <- function(regr){
	regr <- round(regr)
	regr[regr<0] <- 0
	regr[regr>1] <- 1
	return(regr)
}


if(nrow(trainData) <= ncol(trainData)-1){
	stop("Number of rows should be more than number of predictors")
}

# last column is target (numerical, !!!not factor!!! because linear model in this example is suited for regression)
targetName <- colnames(trainData)[ncol(trainData)]
predictorColnames <- colnames(trainData)[-ncol(trainData)]

trainDataWhitenedObj <- whiten(trainData[,predictorColnames])
trainDataColMeans <- colMeans(trainData[,predictorColnames])
trainDataWhitened <- sweep(trainData[,predictorColnames], 2, trainDataColMeans, FUN = "-") %*% trainDataWhitenedObj$whitening
#trainDataWhitened <- trainDataWhitenedObj$U #you can use this too, its already pre-calculated by the model

#amount of created ForeCA components
ForeCA_components <- length(predictorColnames)

forecaObj <- foreca(series = trainDataWhitened, n.comp = ForeCA_components, plot = TRUE)
summary(forecaObj)
plot(forecaObj)

#########################
#Train ForeCA + lm model#
#########################
scoreMatrix <- (trainDataWhitened - forecaObj$center) %*% forecaObj$loadings
#scoreMatrix <- forecaObj$scores #you can use this too, its already pre-calculated by the model
componentNames <- colnames(scoreMatrix)
scoreMatrix <- as.data.frame(scoreMatrix)
scoreMatrix[,targetName] <- trainData[, targetName]

lmModel <- lm(paste(targetName, paste(componentNames,collapse=' + '),sep=' ~ '),data=scoreMatrix)
predictResultsForTrain <- predict(object = lmModel, newdata=scoreMatrix[,-ncol(scoreMatrix)])
predictResultsForTrain <- RegressionToClasses01(predictResultsForTrain)


fronttestDataWhitened <- sweep(trainDataFT[,predictorColnames], 2, trainDataColMeans, FUN = "-") %*% trainDataWhitenedObj$whitening
scoreMatrixFronttest <- (trainDataFT[, predictorColnames] - forecaObj$center) %*% forecaObj$loadings
scoreMatrixFronttest <- as.data.frame(scoreMatrixFronttest)

predictResultsForFronttest <- predict(object = lmModel, newdata=scoreMatrixFronttest)
predictResultsForFronttest <- RegressionToClasses01(predictResultsForFronttest)

"Accuracy on train data using ForeCA + lm"
mean(predictResultsForTrain == trainData[, targetName])
"Accuracy on fronttest data using ForeCA + lm"
mean(predictResultsForFronttest == trainDataFT[, targetName])




################
#Train lm model#
################
rm(lmModel, predictResultsForTrain, predictResultsForFronttest, fronttestDataWhitened, scoreMatrixFronttest)

lmModel <- lm(paste(targetName, paste(predictorColnames,collapse=' + '),sep=' ~ '),data=data.frame(trainData))
predictResultsForTrain <- predict(object = lmModel, newdata=data.frame(trainData[,predictorColnames]))
predictResultsForTrain <- RegressionToClasses01(predictResultsForTrain)

predictResultsForFronttest <- predict(object = lmModel, newdata=data.frame(trainDataFT[,predictorColnames]))
predictResultsForFronttest <- RegressionToClasses01(predictResultsForFronttest)

"Accuracy on train data using only lm"
mean(predictResultsForTrain == trainData[, targetName])
"Accuracy on fronttest data using only lm"
mean(predictResultsForFronttest == trainDataFT[, targetName])






################################################################################
# Train ForeCA + lm model, compare results with different number of components #
#                                                                              #
# it is possible to generate all components,                                   #
# and take less columns from scores and loading matrices                       #
# (number of columns=number of required components).                           #
# Or just execute foreca() again with different component count in n.comp      #
################################################################################
rm(lmModel, predictResultsForTrain, predictResultsForFronttest)
dev.new()

trainingAccuracy <- c()
fronttestAccuracy <- c()
for(i in 1:length(predictorColnames)){
	if(i==1){
		#Linear model does not work with one predictor
		trainingAccuracy <- c(trainingAccuracy, 0)
		fronttestAccuracy <- c(fronttestAccuracy, 0)
		next
	}
	scoreMatrix <- (trainDataWhitened - forecaObj$center) %*% (forecaObj$loadings[,1:i])
	colnames(scoreMatrix) <- colnames(forecaObj$loadings)[1:i]
	componentNames <- colnames(scoreMatrix)
	scoreMatrix <- as.data.frame(scoreMatrix)
	scoreMatrix[,targetName] <- trainData[, targetName]
	lmModel <- lm(paste(targetName, paste(componentNames,collapse=' + '),sep=' ~ '),data=scoreMatrix)
	predictResultsForTrain <- predict(object = lmModel, newdata=scoreMatrix[,-ncol(scoreMatrix)])
	predictResultsForTrain <- RegressionToClasses01(predictResultsForTrain)
	fronttestDataWhitened <- sweep(trainDataFT[,predictorColnames], 2, trainDataColMeans, FUN = "-") %*% trainDataWhitenedObj$whitening
	scoreMatrixFronttest <- (trainDataFT[, predictorColnames] - forecaObj$center) %*% forecaObj$loadings[,1:i]
	colnames(scoreMatrixFronttest) <- colnames(forecaObj$loadings)[1:i]
	scoreMatrixFronttest <- as.data.frame(scoreMatrixFronttest)
	predictResultsForFronttest <- predict(object = lmModel, newdata=scoreMatrixFronttest)
	predictResultsForFronttest <- RegressionToClasses01(predictResultsForFronttest)
	trainingAccuracy <- c(trainingAccuracy, mean(predictResultsForTrain == trainData[, targetName]))
	fronttestAccuracy <- c(fronttestAccuracy, mean(predictResultsForFronttest == trainDataFT[, targetName]))
	plot(trainingAccuracy, type="l", ylab="Accuracy", xlab="Number of components", main="Accuracy of prediction based on components count", col="blue", ylim=c(0,1))
	lines(fronttestAccuracy, type="l", col="red")
	legend(x="bottomright", legend=c("training","fronttest"), fill=c("blue","red"))
	grid()
}