##### 7장 : 신경망(Neural Network)과 서포트 벡터 머신(Support Vector Machines) -------------------

##### Part 1: 신경망 -------------------
## 예제 : 콘크리트의 강도 모델링  ----

## 2단계 : 데이터 준비와 살펴보기 ----
# 데이터 읽기와 구조 확인
concrete <- read.csv("concrete.csv")
str(concrete)

# 정규화 함수
normalize <- function(x) { 
  return((x - min(x)) / (max(x) - min(x)))
}

# 전체 데이터 프레임에 정규화 적용
concrete_norm <- as.data.frame(lapply(concrete, normalize))

# 0과1 사이에 범위 확인
summary(concrete_norm$strength)

# 본래 데이터의 최소값, 최대값 비교
summary(concrete$strength)

# 훈련과 테스트 데이터 생성
concrete_train <- concrete_norm[1:773, ]
concrete_test <- concrete_norm[774:1030, ]

## 3단계 : 데이터로 모델 훈련 ----
# neuralnet 모델 훈련
install.packages("neuralnet")
library(neuralnet)

# 하나의 은닉 뉴런에 대한 단순한 ANN
concrete_model <- neuralnet(formula = strength ~ cement + slag +
                              ash + water + superplastic + 
                              coarseagg + fineagg + age,
                              data = concrete_train)

# 망(network) 시각화
plot(concrete_model)

## 4단계 : 모델 성능 평가 ----

# 모델 결과
model_results <- compute(concrete_model, concrete_test[1:8])

# 강도값 예측
predicted_strength <- model_results$net.result

# 예측값과 실제값간의 상관 관계 확인
cor(predicted_strength, concrete_test$strength)

## 5단계 : 모델 성능 향상 ----
# 5개 은닉 뉴런인 복잡한 뉴런망
concrete_model2 <- neuralnet(strength ~ cement + slag +
                               ash + water + superplastic + 
                               coarseagg + fineagg + age,
                               data = concrete_train, hidden = 5)

# 망(network) 시각화
plot(concrete_model2)

# 결과 평가
model_results2 <- compute(concrete_model2, concrete_test[1:8])
predicted_strength2 <- model_results2$net.result
cor(predicted_strength2, concrete_test$strength)

install.packages("caret")

install.packages("doParallel")

install.packages("kernlab")

install.packages("ggplot2")

install.packages("lattice")

library(ggplot2)

library(lattice)

library(kernlab)

library(caret)

library(doParallel)

# Enable parallel processing.

cl <- makeCluster(detectCores())

registerDoParallel(cl)

# Load the MNIST digit recognition dataset into R

# http://yann.lecun.com/exdb/mnist/

# assume you have all 4 files and gunzip'd them

# creates train$n, train$x, train$y  and test$n, test$x, test$y

# e.g. train$x is a 60000 x 784 matrix, each row is one digit (28x28)

# call:  show_digit(train$x[5,])   to see a digit.

# brendan o'connor - gist.github.com/39760 - anyall.org

load_mnist <- function() {

  load_image_file <- function(filename) {

    ret = list()

    f = file(filename,'rb')

    readBin(f,'integer',n=1,size=4,endian='big')

    ret$n = readBin(f,'integer',n=1,size=4,endian='big')

    nrow = readBin(f,'integer',n=1,size=4,endian='big')

    ncol = readBin(f,'integer',n=1,size=4,endian='big')

    x = readBin(f,'integer',n=ret$n*nrow*ncol,size=1,signed=F)

    ret$x = matrix(x, ncol=nrow*ncol, byrow=T)

    close(f)

    ret

  }

  load_label_file <- function(filename) {

    f = file(filename,'rb')

    readBin(f,'integer',n=1,size=4,endian='big')

    n = readBin(f,'integer',n=1,size=4,endian='big')

    y = readBin(f,'integer',n=n,size=1,signed=F)

    close(f)

    y

  }

  train <<- load_image_file('train-images.idx3-ubyte')

  test <<- load_image_file('t10k-images.idx3-ubyte')

  train$y <<- load_label_file('train-labels.idx1-ubyte')

  test$y <<- load_label_file('t10k-labels.idx1-ubyte')  

}

show_digit <- function(arr784, col=gray(12:1/12), ...) {

  image(matrix(arr784, nrow=28)[,28:1], col=col, ...)

}

train <- data.frame()

test <- data.frame()

# Load data.

load_mnist()

train$x <- train$x / 255

# Setup training data with digit and pixel values with 60/40 split for train/cv.

inTrain = data.frame(y=train$y, train$x)

inTrain$y <- as.factor(inTrain$y)

trainIndex = createDataPartition(inTrain$y, p = 0.60,list=FALSE)

training = inTrain[trainIndex,]

# SVM. 95/94.

fit <- train(y ~ ., data = head(training, 1000), method = 'svmRadial', tuneGrid = data.frame(sigma=0.0107249, C=1))

results

confusionMatrix(results, head(cv$y, 1000))

show_digit(as.matrix(training[5,2:785]))

# Predict the digit.

predict(fit, newdata = training[5,])

# Check the actual answer for the digit.

training[5,1]