AI Archives - 34 중 34 번째 페이지

[PYTHON/TENSORFLOW] 텐서보드 사용하기

■ 텐서보드를 사용하는 방법을 보여준다. 1. 아래 코드를 실행한다. ▶ 예제 코드 (PY)


import tensorflow as tf

a = tf.constant(10, name = "a")
b = tf.constant(90, name = "b")

y = tf.Variable(a + b * 2, name = "y")

model = tf.global_variables_initializer()

with tf.Session() as session:

    merged = tf.summary.merge_all()

    writer = tf.summary.FileWriter("/tmp/tensorflowlogs", session.graph)

    session.run(model)

    print(session.run(y))

import tensorflow as tf

a = tf.constant(10, name = "a")

b = tf.constant(90, name = "b")

y = tf.Variable(a + b * 2, name = "y")

model = tf.global_variables_initializer()

with tf.Session() as session:

merged = tf.summary.merge_all()

writer = tf.summary.FileWriter("/tmp/tensorflowlogs", session.graph)

session.run(model)

print(session.run(y))

2. Anaconda Prompt를 실행하고 아래 명령을 실행한다. ▶

[PYTHON/TENSORFLOW] placeholder 함수 : 변수 만들기

■ placeholder 함수를 사용해 변수를 만드는 방법을 보여준다. ▶ 예제 코드 (PY)


import tensorflow as tf

a = tf.placeholder("int32")
b = tf.placeholder("int32")

y = tf.multiply(a, b)

with tf.Session() as sess:
    result = sess.run(y, feed_dict = { a : 2, b : 5 })
    print(result)

[결과]

10

import tensorflow as tf

a = tf.placeholder("int32")

b = tf.placeholder("int32")

y = tf.multiply(a, b)

with tf.Session() as sess:

result = sess.run(y, feed_dict = { a : 2, b : 5 })

print(result)

[결과]

[PYTHON/TENSORFLOW] constant 함수 : 상수 만들기

■ constant 함수를 사용해 상수를 만드는 방법을 보여준다. ▶ 예제 코드 (PY)


import tensorflow as tf

hello = tf.constant("Hello TensorFlow!")

with tf.Session() as sess:
    result = sess.run(hello)
    print(result)

[결과]

b'Hello TensorFlow!'

import tensorflow as tf

hello = tf.constant("Hello TensorFlow!")

with tf.Session() as sess:

result = sess.run(hello)

print(result)

[결과]

b'Hello TensorFlow!'

[PYTHON/TENSORFLOW] Variable 클래스 : 변수 만들기

■ Variable 클래스를 사용해 변수를 만드는 방법을 보여준다. ▶ 예제 코드 (PY)


import tensorflow as tf

x = tf.constant(1, name = "x")
y = tf.Variable(x + 9, name = "y")

model = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(model)
    result = sess.run(y)
    print(result)

[결과]

10

import tensorflow as tf

x = tf.constant(1, name = "x")

y = tf.Variable(x + 9, name = "y")

model = tf.global_variables_initializer()

with tf.Session() as sess:

sess.run(model)

result = sess.run(y)

print(result)

[결과]

[PYTHON/TENSORFLOW] PYTHON TENSORFLOW 설치하기

■ PYTHON TENSORFLOW를 설치하는 방법을 보여준다. 1. 설치 환경 • 운영 체제 : Microsoft Windows 10 64Bit Version • CUDA 지원 그래픽

[PYTHON/COMMON] 다층 퍼셉트론 신경망 만들기 (MNIST)

■ 다층 퍼셉트론 신경망을 만드는 방법을 보여준다. (MNIST) ▶ main.py


import numpy
import scipy.special

##################################################
# 신경망 클래스
##################################################
class NeuralNetwork:

    ##################################################
    # 생성자
    # inputNodeCount  : 입력 노드 수
    # hiddenNodeCount : 은닉 노드 수
    # outputNodeCount : 출력 노드 수
    # learningRate    : 학습률
    ##################################################
    def __init__(self, inputNodeCount, hiddenNodeCount, outputNodeCount, learningRate):

        self.inputNodeCount  = inputNodeCount
        self.hiddenNodeCount = hiddenNodeCount
        self.outputNodeCount = outputNodeCount

        self.inputHiddenWeightArray  = numpy.random.normal(0.0, pow(self.inputNodeCount , -0.5), \
                                           (self.hiddenNodeCount, self.inputNodeCount ))
        self.hiddenOutputWeightArray = numpy.random.normal(0.0, pow(self.hiddenNodeCount, -0.5), \
                                           (self.outputNodeCount, self.hiddenNodeCount))

        self.learningRate = learningRate

        self.activationFunction = lambda x: scipy.special.expit(x)

        pass

    ##################################################
    # 학습하기
    # sourceInputValueArray  : 소스 입력 값 리스트
    # sourceTargetValueArray : 소스 목표 값 리스트
    ##################################################
    def train(self, sourceInputValueArray, sourceTargetValueArray):

        inputValueArray  = numpy.array(sourceInputValueArray , ndmin = 2).T
        targetValueArray = numpy.array(sourceTargetValueArray, ndmin = 2).T

        # 은닉 계층으로 들어오는 신호를 계산한다.
        hiddenNodeInputValueArray = numpy.dot(self.inputHiddenWeightArray, inputValueArray)

        # 은닉 계층에서 나가는 신호를 계산한다.
        hiddenNodeOutputValueArray = self.activationFunction(hiddenNodeInputValueArray)

        # 출력 계층으로 들어오는 신호를 계산한다.
        outputNodeInputValueArray = numpy.dot(self.hiddenOutputWeightArray, hiddenNodeOutputValueArray)

        # 출력 계층에서 나가는 신호를 계산한다.
        outputNodeOutputValueArray = self.activationFunction(outputNodeInputValueArray)

        # 출력 계층의 오차를 계산한다. (실제 값 - 계산 값)
        outputNodeErrorArray = targetValueArray - outputNodeOutputValueArray

        # 은닉 계층의 오차는 가중치에 의해 나뉜 출력 계층의 오차들을 재조합해 계산한다.
        hiddenNodeErrorArray = numpy.dot(self.hiddenOutputWeightArray.T, outputNodeErrorArray)

        # 은닉 계층과 출력 계층 간의 가중치 오차를 업데이트 한다.
        self.hiddenOutputWeightArray += self.learningRate * \
            numpy.dot((outputNodeErrorArray * outputNodeOutputValueArray * (1.0 - outputNodeOutputValueArray)), \
                numpy.transpose(hiddenNodeOutputValueArray))

        # 입력 계층과 은닉 계층 간의 가중치 오차를 업데이트 한다.
        self.inputHiddenWeightArray += self.learningRate * \
            numpy.dot((hiddenNodeErrorArray * hiddenNodeOutputValueArray * (1.0 - hiddenNodeOutputValueArray)), \
                numpy.transpose(inputValueArray))

        pass

    ##################################################
    # 질의하기
    # sourceInputValueArray : 소스 입력 값 리스트
    ##################################################
    def query(self, sourceInputValueArray):

        inputValueArray = numpy.array(sourceInputValueArray, ndmin = 2).T

        # 은닉 계층으로 들어오는 신호를 계산한다.
        hiddenNodeInputValueArray = numpy.dot(self.inputHiddenWeightArray, inputValueArray)

        # 은닉 계층에서 나가는 신호를 계산한다.
        hiddenNodeOutputValueArray = self.activationFunction(hiddenNodeInputValueArray)

        # 출력 계층으로 들어오는 신호를 계산한다.
        outputNodeInputValueArray = numpy.dot(self.hiddenOutputWeightArray, hiddenNodeOutputValueArray)

        # 출력 계층에서 나가는 신호를 계산한다.
        outputNodeOutputValueArray = self.activationFunction(outputNodeInputValueArray)

        return outputNodeOutputValueArray

        pass

##################################################
# 신경망 객체를 생성한다.
##################################################
print("신경망 객체를 생성한다.")

inputNodeCount  = 784
hiddenNodeCount = 100
outputNodeCount = 10

learningRate = 0.1

neuralNetwork = NeuralNetwork(inputNodeCount, hiddenNodeCount, outputNodeCount, learningRate)

##################################################
# MNIST 학습 데이터를 로드한다.
##################################################
print("MNIST 학습 데이터를 로드한다.")

trainingDataFile = open("mnist_train.csv", 'r')

trainingDataList = trainingDataFile.readlines()

trainingDataFile.close()

print("학습 데이터 : %s건" % len(trainingDataList))

##################################################
# MNIST 학습 데이터를 사용해 신경망을 학습시킨다.
##################################################
print("MNIST 학습 데이터를 사용해 신경망을 학습시킨다.")

epochCount = 5

for epoch in range(epochCount):

    print("학습 주기 = ", epoch)

    for trainingData in trainingDataList:

        trainigDataArray = trainingData.split(',')

        # 학습에 사용되는 MNIST 숫자 이미지는 28×28 픽셀 = 784개 데이터를 갖으며 훈련 데이터 배열에서 두번째 컬럼 이후 부터 저장된다.
        testInputValueArray = (numpy.asfarray(trainigDataArray[1:]) / 255.0 * 0.99) + 0.01

        trainingTargetValueArray = numpy.zeros(outputNodeCount) + 0.01

        trainingTargetValueArray[int(trainigDataArray[0])] = 0.99

        neuralNetwork.train(testInputValueArray, trainingTargetValueArray)

        pass

    pass

print("학습 완료")

##################################################
# MNIST 테스트 데이터를 로드한다.
##################################################
print("MNIST 테스트 데이터를 로드한다.")

testDataFile = open("mnist_test.csv", 'r')

testDataList = testDataFile.readlines()

testDataFile.close()

print("테스트 데이터 : %s건" % len(testDataList))

##################################################
# MNIST 테스트 데이터를 사용해 신경망을 테스트한다.
##################################################
print("MNIST 테스트 데이터를 사용해 신경망을 테스트한다.")

scoreCardList = []

for testData in testDataList:

    testDataArray = testData.split(',')

    correctLabel = int(testDataArray[0])

    testInputValueArray = (numpy.asfarray(testDataArray[1:]) / 255.0 * 0.99) + 0.01

    testOutputValueArray = neuralNetwork.query(testInputValueArray)

    answerLabel = numpy.argmax(testOutputValueArray)

    if(answerLabel == correctLabel):

        scoreCardList.append(1)

        pass

    else:

        scoreCardList.append(0)

        pass

    pass

scoreCardArray = numpy.asarray(scoreCardList)

print("정확도 = ", scoreCardArray.sum() / scoreCardArray.size)

##################################################
# MNIST 테스트 데이터 1건을 사용해 신경망을 테스트한다.
##################################################

import matplotlib.pyplot as pp

testDataArray = testDataList[200].split(',')

print("테스트 데이터 정답 : ", testDataArray[0])

testInputValueArray = (numpy.asfarray(testDataArray[1:]) / 255.0 * 0.99) + 0.01

testOutputValueArray = neuralNetwork.query(testInputValueArray)

print("테스트 결과 값 : ", numpy.argmax(testOutputValueArray))
print()
print(testOutputValueArray)

testDataImageArray = numpy.asfarray(testDataArray[1:]).reshape((28, 28))

pp.imshow(testDataImageArray, cmap='Greys', interpolation='None')

pp.show()

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

import numpy

import scipy.special

##################################################

# 신경망 클래스

##################################################

class NeuralNetwork:

##################################################

# 생성자

# inputNodeCount : 입력 노드 수

# hiddenNodeCount : 은닉 노드 수

# outputNodeCount : 출력 노드 수

# learningRate : 학습률

##################################################

def __init__(self, inputNodeCount, hiddenNodeCount, outputNodeCount, learningRate):

self.inputNodeCount = inputNodeCount

self.hiddenNodeCount = hiddenNodeCount

self.outputNodeCount = outputNodeCount

self.inputHiddenWeightArray = numpy.random.normal(0.0, pow(self.inputNodeCount , -0.5), \

(self.hiddenNodeCount, self.inputNodeCount ))

self.hiddenOutputWeightArray = numpy.random.normal(0.0, pow(self.hiddenNodeCount, -0.5), \

(self.outputNodeCount, self.hiddenNodeCount))

self.learningRate = learningRate

self.activationFunction = lambda x: scipy.special.expit(x)

pass

##################################################

# 학습하기

# sourceInputValueArray : 소스 입력 값 리스트

# sourceTargetValueArray : 소스 목표 값 리스트

##################################################

def train(self, sourceInputValueArray, sourceTargetValueArray):

inputValueArray = numpy.array(sourceInputValueArray , ndmin = 2).T

targetValueArray = numpy.array(sourceTargetValueArray, ndmin = 2).T

# 은닉 계층으로 들어오는 신호를 계산한다.

hiddenNodeInputValueArray = numpy.dot(self.inputHiddenWeightArray, inputValueArray)

# 은닉 계층에서 나가는 신호를 계산한다.

hiddenNodeOutputValueArray = self.activationFunction(hiddenNodeInputValueArray)

# 출력 계층으로 들어오는 신호를 계산한다.

outputNodeInputValueArray = numpy.dot(self.hiddenOutputWeightArray, hiddenNodeOutputValueArray)

# 출력 계층에서 나가는 신호를 계산한다.

outputNodeOutputValueArray = self.activationFunction(outputNodeInputValueArray)

# 출력 계층의 오차를 계산한다. (실제 값 - 계산 값)

outputNodeErrorArray = targetValueArray - outputNodeOutputValueArray

# 은닉 계층의 오차는 가중치에 의해 나뉜 출력 계층의 오차들을 재조합해 계산한다.

hiddenNodeErrorArray = numpy.dot(self.hiddenOutputWeightArray.T, outputNodeErrorArray)

# 은닉 계층과 출력 계층 간의 가중치 오차를 업데이트 한다.

self.hiddenOutputWeightArray += self.learningRate * \

numpy.dot((outputNodeErrorArray * outputNodeOutputValueArray * (1.0 - outputNodeOutputValueArray)), \

numpy.transpose(hiddenNodeOutputValueArray))

# 입력 계층과 은닉 계층 간의 가중치 오차를 업데이트 한다.

self.inputHiddenWeightArray += self.learningRate * \

numpy.dot((hiddenNodeErrorArray * hiddenNodeOutputValueArray * (1.0 - hiddenNodeOutputValueArray)), \

numpy.transpose(inputValueArray))

pass

##################################################

# 질의하기

# sourceInputValueArray : 소스 입력 값 리스트

##################################################

def query(self, sourceInputValueArray):

inputValueArray = numpy.array(sourceInputValueArray, ndmin = 2).T

# 은닉 계층으로 들어오는 신호를 계산한다.

hiddenNodeInputValueArray = numpy.dot(self.inputHiddenWeightArray, inputValueArray)

# 은닉 계층에서 나가는 신호를 계산한다.

hiddenNodeOutputValueArray = self.activationFunction(hiddenNodeInputValueArray)

# 출력 계층으로 들어오는 신호를 계산한다.

outputNodeInputValueArray = numpy.dot(self.hiddenOutputWeightArray, hiddenNodeOutputValueArray)

# 출력 계층에서 나가는 신호를 계산한다.

outputNodeOutputValueArray = self.activationFunction(outputNodeInputValueArray)

return outputNodeOutputValueArray

pass

##################################################

# 신경망 객체를 생성한다.

##################################################

print("신경망 객체를 생성한다.")

inputNodeCount = 784

hiddenNodeCount = 100

outputNodeCount = 10

learningRate = 0.1

neuralNetwork = NeuralNetwork(inputNodeCount, hiddenNodeCount, outputNodeCount, learningRate)

##################################################

# MNIST 학습 데이터를 로드한다.

##################################################

print("MNIST 학습 데이터를 로드한다.")

trainingDataFile = open("mnist_train.csv", 'r')

trainingDataList = trainingDataFile.readlines()

trainingDataFile.close()

print("학습 데이터 : %s건" % len(trainingDataList))

##################################################

# MNIST 학습 데이터를 사용해 신경망을 학습시킨다.

##################################################

print("MNIST 학습 데이터를 사용해 신경망을 학습시킨다.")

epochCount = 5

for epoch in range(epochCount):

print("학습 주기 = ", epoch)

for trainingData in trainingDataList:

trainigDataArray = trainingData.split(',')

# 학습에 사용되는 MNIST 숫자 이미지는 28×28 픽셀 = 784개 데이터를 갖으며 훈련 데이터 배열에서 두번째 컬럼 이후 부터 저장된다.

testInputValueArray = (numpy.asfarray(trainigDataArray[1:]) / 255.0 * 0.99) + 0.01

trainingTargetValueArray = numpy.zeros(outputNodeCount) + 0.01

trainingTargetValueArray[int(trainigDataArray[0])] = 0.99

neuralNetwork.train(testInputValueArray, trainingTargetValueArray)

pass

print("학습 완료")

##################################################

# MNIST 테스트 데이터를 로드한다.

##################################################

print("MNIST 테스트 데이터를 로드한다.")

testDataFile = open("mnist_test.csv", 'r')

testDataList = testDataFile.readlines()

testDataFile.close()

print("테스트 데이터 : %s건" % len(testDataList))

##################################################

# MNIST 테스트 데이터를 사용해 신경망을 테스트한다.

##################################################

print("MNIST 테스트 데이터를 사용해 신경망을 테스트한다.")

scoreCardList = []

for testData in testDataList:

testDataArray = testData.split(',')

correctLabel = int(testDataArray[0])

testInputValueArray = (numpy.asfarray(testDataArray[1:]) / 255.0 * 0.99) + 0.01

testOutputValueArray = neuralNetwork.query(testInputValueArray)

answerLabel = numpy.argmax(testOutputValueArray)

if(answerLabel == correctLabel):

scoreCardList.append(1)

pass

else:

scoreCardList.append(0)

pass

scoreCardArray = numpy.asarray(scoreCardList)

print("정확도 = ", scoreCardArray.sum() / scoreCardArray.size)

##################################################

# MNIST 테스트 데이터 1건을 사용해 신경망을 테스트한다.

##################################################

import matplotlib.pyplot as pp

testDataArray = testDataList[200].split(',')

print("테스트 데이터 정답 : ", testDataArray[0])

testInputValueArray = (numpy.asfarray(testDataArray[1:]) / 255.0 * 0.99) + 0.01

testOutputValueArray = neuralNetwork.query(testInputValueArray)

print("테스트 결과 값 : ", numpy.argmax(testOutputValueArray))

print()

print(testOutputValueArray)

testDataImageArray = numpy.asfarray(testDataArray[1:]).reshape((28, 28))

pp.imshow(testDataImageArray, cmap='Greys', interpolation='None')

pp.show()

▶ requirements.txt


cycler==0.11.0
fonttools==4.37.1
kiwisolver==1.4.4
matplotlib==3.5.3
numpy==1.23.2
packaging==21.3
Pillow==9.2.0
pip==22.0.4
pyparsing==3.0.9
python-dateutil==2.8.2
scipy==1.9.0
setuptools==58.1.0
six==1.16.0

cycler==0.11.0

fonttools==4.37.1

kiwisolver==1.4.4

matplotlib==3.5.3

numpy==1.23.2

packaging==21.3

Pillow==9.2.0

pip==22.0.4

pyparsing==3.0.9

python-dateutil==2.8.2

scipy==1.9.0

setuptools==58.1.0

six==1.16.0

mnist.zip

[C#/COMMON] 심층 신경망(Deep Neural Network) 알고리즘 사용하기

■ 심층 신경망(Deep Neural Network) 알고리즘을 사용하는 방법을 보여준다. ▶ DeepNeuralNetwork.cs


using System;

namespace TestProject
{
    /// <summary>
    /// 심층 신경망
    /// </summary>
    public class DeepNeuralNetwork
    {
        //////////////////////////////////////////////////////////////////////////////////////////////////// Field
        ////////////////////////////////////////////////////////////////////////////////////////// Static
        //////////////////////////////////////////////////////////////////////////////// Private

        #region Field

        /// <summary>
        /// 난수기
        /// </summary>
        private static Random _random;

        #endregion

        ////////////////////////////////////////////////////////////////////////////////////////// Instance
        //////////////////////////////////////////////////////////////////////////////// Private

        #region Field

        /// <summary>
        /// 입력 노드 수
        /// </summary>
        private int inputNodeCount;

        /// <summary>
        /// 은닉 노드 수 배열
        /// </summary>
        private int[] hiddenNodeCountArray;

        /// <summary>
        /// 출력 노드 수
        /// </summary>
        private int outputNodeCount;

        /// <summary>
        /// 은닉층 수
        /// </summary>
        private int hiddenLayerCount;

        /// <summary>
        /// 입력 노드 배열
        /// </summary>
        private double[] inputNodeArray;

        /// <summary>
        /// 은닉 노드 배열
        /// </summary>
        private double[][] hiddenNodeArray;

        /// <summary>
        /// 출력 노드 배열
        /// </summary>
        private double[] outputNodeArray;

        /// <summary>
        /// 입력/은닉 가중치 배열
        /// </summary>
        private double[][] inputHiddenWeightArray;

        /// <summary>
        /// 은닉/은닉 가중치 배열
        /// </summary>
        private double[][][] hiddenHiddenWeightArray;

        /// <summary>
        /// 은닉/출력 가중치 배열
        /// </summary>
        private double[][] hiddenOutputWeightArray;

        /// <summary>
        /// 은닉 노드 바이어스 배열
        /// </summary>
        private double[][] hiddenNodeBiasArray;

        /// <summary>
        /// 출력 노드 바이어스 배열
        /// </summary>
        private double[] outputNodeBiasArray;

        /// <summary>
        /// 입력/은닉 1개 그라디언트
        /// </summary>
        private double inputHiddenOneGradient;

        #endregion

        //////////////////////////////////////////////////////////////////////////////////////////////////// Constructor
        ////////////////////////////////////////////////////////////////////////////////////////// Public

        #region 생성자 - DeepNeuralNetwork(inputNodeCount, hiddenNodeCountArray, outputNodeCount)

        /// <summary>
        /// 생성자
        /// </summary>
        /// <param name="inputNodeCount">입력 노드 수</param>
        /// <param name="hiddenNodeCountArray">은닉 노드 수 배열</param>
        /// <param name="outputNodeCount">출력 노드 수</param>
        public DeepNeuralNetwork(int inputNodeCount, int[] hiddenNodeCountArray, int outputNodeCount)
        {
            _random = new Random(0);

            this.inputNodeCount = inputNodeCount;

            this.hiddenNodeCountArray = new int[hiddenNodeCountArray.Length];

            for(int i = 0; i < hiddenNodeCountArray.Length; ++i)
            {
                this.hiddenNodeCountArray[i] = hiddenNodeCountArray[i];
            }

            this.outputNodeCount = outputNodeCount;

            this.hiddenLayerCount = hiddenNodeCountArray.Length;

            this.inputNodeArray  = new double[inputNodeCount];
            this.hiddenNodeArray = GetJaggedMatrix(hiddenNodeCountArray);
            this.outputNodeArray = new double[outputNodeCount];

            this.inputHiddenWeightArray  = GetMatrix(inputNodeCount, hiddenNodeCountArray[0]);
            this.hiddenOutputWeightArray = GetMatrix(hiddenNodeCountArray[this.hiddenLayerCount - 1], outputNodeCount);

            this.hiddenHiddenWeightArray = new double[this.hiddenLayerCount - 1][][];

            for(int i = 0; i < this.hiddenHiddenWeightArray.Length; ++i)
            {
                int rowCount    = hiddenNodeCountArray[i    ];
                int columnCount = hiddenNodeCountArray[i + 1];

                this.hiddenHiddenWeightArray[i] = GetMatrix(rowCount, columnCount);
            }

            this.hiddenNodeBiasArray = GetJaggedMatrix(hiddenNodeCountArray);
            this.outputNodeBiasArray = new double[outputNodeCount];

            InitializeWeightArray();
        }

        #endregion

        //////////////////////////////////////////////////////////////////////////////////////////////////// Method
        ////////////////////////////////////////////////////////////////////////////////////////// Static
        //////////////////////////////////////////////////////////////////////////////// Public

        #region 전체 가중치 수 구하기 - GetTotalWeightCount(inputNodeCount, hiddenNodeCountArray, outputNodeCount)

        /// <summary>
        /// 전체 가중치 수 구하기
        /// </summary>
        /// <param name="inputNodeCount">입력 노드 수</param>
        /// <param name="hiddenNodeCountArray">은닉 노드 수 배열</param>
        /// <param name="outputNodeCount">출력 노드 수</param>
        /// <returns>전체 가중치 수</returns>
        public static int GetTotalWeightCount(int inputNodeCount, int[] hiddenNodeCountArray, int outputNodeCount)
        {
            int inputHeightWeightCount = inputNodeCount * hiddenNodeCountArray[0];

            int hiddenNodeCountArrayLength = hiddenNodeCountArray.Length;

            int hiddenHiddenWeightCount = 0;

            for(int i = 0; i < hiddenNodeCountArrayLength - 1; ++i)
            {
                int rowCount    = hiddenNodeCountArray[i    ];
                int columnCount = hiddenNodeCountArray[i + 1];

                hiddenHiddenWeightCount += rowCount * columnCount;
            }

            int hiddenOutputWeightCount = hiddenNodeCountArray[hiddenNodeCountArrayLength - 1] * outputNodeCount;

            int hiddenBiasCount = 0;

            for(int i = 0; i < hiddenNodeCountArrayLength; ++i)
            {
                hiddenBiasCount += hiddenNodeCountArray[i];
            }

            int outputBiasCount = outputNodeCount;

            int totalWeightCount = inputHeightWeightCount + hiddenHiddenWeightCount + hiddenOutputWeightCount + hiddenBiasCount + outputBiasCount;

            return totalWeightCount;
        }

        #endregion

        //////////////////////////////////////////////////////////////////////////////// Private

        #region 배열 표시하기 - DisplayArray(sourceArray, decimalPointCount)

        /// <summary>
        /// 배열 표시하기
        /// </summary>
        /// <param name="sourceArray">소스 배열</param>
        /// <param name="decimalPointCount">소수점 자리수</param>
        private static void DisplayArray(double[] sourceArray, int decimalPointCount)
        {
            string displayFormat = string.Format("F{0}", decimalPointCount);

            for(int i = 0; i < sourceArray.Length; ++i)
            {
                Console.Write(sourceArray[i].ToString(displayFormat) + " ");
            }

            Console.WriteLine();
        }

        #endregion
        #region JAGGED 매트릭스 구하기 - GetJaggedMatrix(columnCountArray)

        /// <summary>
        /// JAGGED 매트릭스 구하기
        /// </summary>
        /// <param name="columnCountArray">컬럼 수 배열</param>
        /// <returns>JAGGED 매트릭스</returns>
        private static double[][] GetJaggedMatrix(int[] columnCountArray)
        {
            int rowCount = columnCountArray.Length;

            double[][] resultArray = new double[rowCount][];

            for(int i = 0; i < rowCount; ++i)
            {
                int columnCount = columnCountArray[i];

                resultArray[i] = new double[columnCount];
            }

            return resultArray;
        }

        #endregion
        #region 매트릭스 구하기 - GetMatrix(rowCount, columnCount)

        /// <summary>
        /// 매트릭스 구하기
        /// </summary>
        /// <param name="rowCount">행 수</param>
        /// <param name="columnCount">컬럼 수</param>
        /// <returns>매트릭스</returns>
        private static double[][] GetMatrix(int rowCount, int columnCount)
        {
            double[][] resultArray = new double[rowCount][];

            for(int i = 0; i < rowCount; ++i)
            {
                resultArray[i] = new double[columnCount];
            }

            return resultArray;
        }

        #endregion
        #region 최대 값 인덱스 구하기 - GetMaximumValueIndex(sourceArray)

        /// <summary>
        /// 최대 값 인덱스 구하기
        /// </summary>
        /// <param name="sourceArray">소스 배열</param>
        /// <returns>최대 값 인덱스</returns>
        private static int GetMaximumValueIndex(double[] sourceArray)
        {
            int    maximumValueIndex = 0;
            double maximumValue = sourceArray[0];
            int    sourceArrayLength = sourceArray.Length;

            for(int i = 0; i < sourceArrayLength; ++i)
            {
                if(sourceArray[i] > maximumValue)
                {
                    maximumValue = sourceArray[i];

                    maximumValueIndex = i;
                }
            }

            return maximumValueIndex;
        }

        #endregion
        #region 하이퍼볼릭 탄젠트 구하기 - GetHyperbolicTangent(x)

        /// <summary>
        /// 하이퍼볼릭 탄젠트 구하기
        /// </summary>
        /// <param name="x">X</param>
        /// <returns>하이퍼볼릭 탄젠트</returns>
        private static double GetHyperbolicTangent(double x)
        {
            if(x < -20d)
            {
                return -1d;
            }
            else if(x > 20d)
            {
                return 1d;
            }
            else
            {
                return Math.Tanh(x);
            }
        }

        #endregion
        #region 소프트맥스 구하기 - GetSoftmax(sourceArray)

        /// <summary>
        /// 소프트맥스 구하기
        /// </summary>
        /// <param name="sourceArray">소스 배열</param>
        /// <returns>소프트맥스</returns>
        private static double[] GetSoftmax(double[] sourceArray)
        {
            int sourceArrayLength = sourceArray.Length;

            double summary = 0d;

            for(int i = 0; i < sourceArrayLength; ++i)
            {
                summary += Math.Exp(sourceArray[i]);
            }

            double[] resultArray = new double[sourceArrayLength];

            for(int i = 0; i < sourceArrayLength; ++i)
            {
                resultArray[i] = Math.Exp(sourceArray[i]) / summary;
            }

            return resultArray;
        }

        #endregion

        ////////////////////////////////////////////////////////////////////////////////////////// Instance
        //////////////////////////////////////////////////////////////////////////////// Public

        #region 가중치 배열 설정하기 - SetWeightArray(sourceArray)

        /// <summary>
        /// 가중치 배열 설정하기
        /// </summary>
        /// <param name="sourceArray">소스 배열</param>
        public void SetWeightArray(double[] sourceArray)
        {
            int totalWeightCount = GetTotalWeightCount(this.inputNodeCount, this.hiddenNodeCountArray, this.outputNodeCount);

            if(sourceArray.Length != totalWeightCount)
            {
                throw new Exception("부적절한 가중치 배열 길이 입니다.");
            }

            int firstHiddenNodeArrayLength = this.hiddenNodeArray[0].Length;

            int weightArrayIndex = 0;

            for(int i = 0; i < this.inputNodeCount; ++i)
            {
                for(int j = 0; j < firstHiddenNodeArrayLength; ++j)
                {
                    this.inputHiddenWeightArray[i][j] = sourceArray[weightArrayIndex++];
                }
            }

            for(int i = 0; i < this.hiddenLayerCount - 1; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    for(int k = 0; k < this.hiddenNodeCountArray[i + 1]; ++k)
                    {
                        this.hiddenHiddenWeightArray[i][j][k] = sourceArray[weightArrayIndex++];
                    }
                }
            }

            int hiddenNodeCountArrayLastIndex = this.hiddenLayerCount - 1;

            for(int i = 0; i < this.hiddenNodeCountArray[hiddenNodeCountArrayLastIndex]; ++i)
            {
                for(int j = 0; j < this.outputNodeCount; ++j)
                {
                    this.hiddenOutputWeightArray[i][j] = sourceArray[weightArrayIndex++];
                }
            }

            for(int i = 0; i < this.hiddenLayerCount; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    this.hiddenNodeBiasArray[i][j] = sourceArray[weightArrayIndex++];
                }
            }

            for(int i = 0; i < this.outputNodeCount; ++i)
            {
                this.outputNodeBiasArray[i] = sourceArray[weightArrayIndex++];
            }
        }

        #endregion
        #region 질의하기 - Query(inputValueArray)

        /// <summary>
        /// 질의하기
        /// </summary>
        /// <param name="inputValueArray">입력 값 배열</param>
        /// <returns>출력 값 배열</returns>
        public double[] Query(double[] inputValueArray)
        {
            for(int i = 0; i < this.inputNodeCount; ++i)
            {
                this.inputNodeArray[i] = inputValueArray[i];
            }

            for(int i = 0; i < this.hiddenLayerCount; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    this.hiddenNodeArray[i][j] = 0d;
                }
            }

            for(int i = 0; i < this.outputNodeCount; ++i)
            {
                this.outputNodeArray[i] = 0.0;
            }

            for(int i = 0; i < this.hiddenNodeCountArray[0]; ++i)
            {
                for(int j = 0; j < this.inputNodeCount; ++j)
                {
                    this.hiddenNodeArray[0][i] += this.inputHiddenWeightArray[j][i] * this.inputNodeArray[j];
                }

                this.hiddenNodeArray[0][i] += this.hiddenNodeBiasArray[0][i];

                this.hiddenNodeArray[0][i] = Math.Tanh(this.hiddenNodeArray[0][i]);
            }

            for(int i = 1; i < this.hiddenLayerCount; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    for(int k = 0; k < this.hiddenNodeCountArray[i - 1]; ++k)
                    {
                        this.hiddenNodeArray[i][j] += this.hiddenHiddenWeightArray[i - 1][k][j] * this.hiddenNodeArray[i - 1][k];
                    }

                    this.hiddenNodeArray[i][j] += this.hiddenNodeBiasArray[i][j];

                    this.hiddenNodeArray[i][j] = Math.Tanh(this.hiddenNodeArray[i][j]);
                }
            }

            for(int i = 0; i < this.outputNodeCount; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[this.hiddenLayerCount - 1]; ++j)
                {
                    this.outputNodeArray[i] += this.hiddenOutputWeightArray[j][i] * this.hiddenNodeArray[this.hiddenLayerCount - 1][j];
                }

                this.outputNodeArray[i] += this.outputNodeBiasArray[i];
            }

            double[] outputArray = GetSoftmax(this.outputNodeArray);

            for(int k = 0; k < this.outputNodeCount; ++k)
            {
                this.outputNodeArray[k] = outputArray[k];
            }

            return outputArray;
        }

        #endregion
        #region 오차 구하기 - GetError(trainingDataArray, verbose)

        /// <summary>
        /// 오차 구하기
        /// </summary>
        /// <param name="trainingDataArray">훈련 데이터 배열</param>
        /// <param name="verbose">서술 여부</param>
        /// <returns>오차</returns>
        public double GetError(double[][] trainingDataArray, bool verbose)
        {
            double   summarySquaredError = 0d;
            double[] inputValueArray     = new double[this.inputNodeCount ];
            double[] targetValueArray    = new double[this.outputNodeCount];

            for(int i = 0; i < trainingDataArray.Length; ++i)
            {
                Array.Copy(trainingDataArray[i], inputValueArray, this.inputNodeCount);

                Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                double[] outputValueArray = this.Query(inputValueArray);

                if(verbose == true)
                {
                    DisplayArray(outputValueArray, 4);
                    DisplayArray(targetValueArray, 4);

                    Console.WriteLine();

                    Console.ReadLine();
                }

                for(int j = 0; j < this.outputNodeCount; ++j)
                {
                    double error = targetValueArray[j] - outputValueArray[j];

                    summarySquaredError += error * error;
                }
            }

            return summarySquaredError / (trainingDataArray.Length * this.outputNodeCount);
        }

        #endregion
        #region 오차 구하기 - GetError(trainingDataArray, weightArray)

        /// <summary>
        /// 오차 구하기
        /// </summary>
        /// <param name="trainingDataArray">훈련 데이터 배열</param>
        /// <param name="weightArray">가중치 배열</param>
        /// <returns>오차</returns>
        public double GetError(double[][] trainingDataArray, double[] weightArray)
        {
            SetWeightArray(weightArray);

            double   summarySquaredError = 0d;
            double[] intputValueArray    = new double[this.inputNodeCount ];
            double[] targetValueArray    = new double[this.outputNodeCount];

            int trainingDataArrayLength = trainingDataArray.Length;

            for(int i = 0; i < trainingDataArrayLength; ++i)
            {
                Array.Copy(trainingDataArray[i], intputValueArray, this.inputNodeCount);

                Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                double[] outputValueArray = Query(intputValueArray);

                for(int j = 0; j < this.outputNodeCount; ++j)
                {
                    double error = targetValueArray[j] - outputValueArray[j];

                    summarySquaredError += error * error;
                }
            }

            return summarySquaredError / trainingDataArrayLength;
        }

        #endregion
        #region 정확도 구하기 - GetAccuracy(trainingDataArray, verbose)

        /// <summary>
        /// 정확도 구하기
        /// </summary>
        /// <param name="trainingDataArray">훈련 데이터 배열</param>
        /// <param name="verbose">서술 여부</param>
        /// <returns>정확도</returns>
        public double GetAccuracy(double[][] trainingDataArray, bool verbose)
        {
            int      correctCount = 0;
            int      wrongCount = 0;
            double[] intputValueArray = new double[this.inputNodeCount ];
            double[] targetValueArray = new double[this.outputNodeCount];
            double[] outputValueArray;

            int trainingDataArrayLength = trainingDataArray.Length;

            for(int i = 0; i < trainingDataArrayLength; ++i)
            {
                Array.Copy(trainingDataArray[i], intputValueArray, this.inputNodeCount);

                Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                outputValueArray = Query(intputValueArray);

                if(verbose == true)
                {
                    DisplayArray(outputValueArray, 4);
                    DisplayArray(targetValueArray, 4);

                    Console.WriteLine();
                    Console.ReadLine();
                }

                int outputMaximumValueIndex = GetMaximumValueIndex(outputValueArray);
                int targetMaximumValueIndex = GetMaximumValueIndex(targetValueArray);

                if(outputMaximumValueIndex == targetMaximumValueIndex)
                {
                    ++correctCount;
                }
                else
                {
                    ++wrongCount;
                }
            }

            return (correctCount * 1d) / (correctCount + wrongCount);
        }

        #endregion
        #region 정확도 구하기 - GetAccuracy(trainingDataArray, weightArray)

        /// <summary>
        /// 정확도 구하기
        /// </summary>
        /// <param name="trainingDataArray">훈련 데이터 배열</param>
        /// <param name="weightArray">가중치 배열</param>
        /// <returns>정확도</returns>
        public double GetAccuracy(double[][] trainingDataArray, double[] weightArray)
        {
            SetWeightArray(weightArray);

            int      correctCount     = 0;
            int      wrongCount       = 0;
            double[] inputValueArray  = new double[this.inputNodeCount ];
            double[] targetValueArray = new double[this.outputNodeCount];
            double[] outputValueArray;

            int trainingDataArrayLength = trainingDataArray.Length;

            for(int i = 0; i < trainingDataArrayLength; ++i)
            {
                Array.Copy(trainingDataArray[i], inputValueArray, this.inputNodeCount);

                Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                outputValueArray = Query(inputValueArray);

                int outputMaximumValueIndex = GetMaximumValueIndex(outputValueArray);
                int targetMaximumValueIndex = GetMaximumValueIndex(targetValueArray);

                if(outputMaximumValueIndex == targetMaximumValueIndex)
                {
                    ++correctCount;
                }
                else
                {
                    ++wrongCount;
                }
            }

            return (correctCount * 1d) / (correctCount + wrongCount);
        }

        #endregion
        #region 가중치 배열 구하기 - GetWeightArray()

        /// <summary>
        /// 가중치 배열 구하기
        /// </summary>
        /// <returns>가중치 배열</returns>
        public double[] GetWeightArray()
        {
            int totalWeightCount = GetTotalWeightCount(this.inputNodeCount, this.hiddenNodeCountArray, this.outputNodeCount);

            double[] resultArray = new double[totalWeightCount];

            int index = 0;

            for(int i = 0; i < this.inputNodeCount; ++i)
            {
                for(int j = 0; j < this.hiddenNodeArray[0].Length; ++j)
                {
                    resultArray[index++] = this.inputHiddenWeightArray[i][j];
                }
            }

            for(int i = 0; i < this.hiddenLayerCount - 1; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    for(int k = 0; k < this.hiddenNodeCountArray[i + 1]; ++k)
                    {
                        resultArray[index++] = this.hiddenHiddenWeightArray[i][j][k];
                    }
                }
            }

            int hiddenNodeCountArrayLastIndex = this.hiddenLayerCount - 1;

            for(int i = 0; i < this.hiddenNodeCountArray[hiddenNodeCountArrayLastIndex]; ++i)
            {
                for(int j = 0; j < this.outputNodeCount; ++j)
                {
                    resultArray[index++] = this.hiddenOutputWeightArray[i][j];
                }
            }

            for(int i = 0; i < this.hiddenLayerCount; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    resultArray[index++] = this.hiddenNodeBiasArray[i][j];
                }
            }

            for(int i = 0; i < this.outputNodeCount; ++i)
            {
                resultArray[index++] = this.outputNodeBiasArray[i];
            }

            return resultArray;
        }

        #endregion
        #region 훈련하기 - Train(trainingDataArray, maximumEpochCount, learningRate, momentum, showErrorCount)

        /// <summary>
        /// 훈련하기
        /// </summary>
        /// <param name="trainingDataArray">훈련 데이터 배열</param>
        /// <param name="maximumEpochCount">최대 반복 차수</param>
        /// <param name="learningRate">학습률</param>
        /// <param name="momentum">모멘텀</param>
        /// <param name="showErrorCount">오차 표시 수</param>
        /// <returns>최적 가중치 배열</returns>
        public double[] Train(double[][] trainingDataArray, int maximumEpochCount, double learningRate, double momentum, int showErrorCount)
        {
            double[][]   hiddenOutputGradientArray = GetMatrix(this.hiddenNodeCountArray[this.hiddenLayerCount - 1], this.outputNodeCount);
            double[][][] hiddenHiddenGradientArray = new double[this.hiddenLayerCount - 1][][];

            for(int i = 0; i < hiddenHiddenGradientArray.Length; ++i)
            {
                int rowCount    = this.hiddenNodeCountArray[i    ];
                int columnCount = this.hiddenNodeCountArray[i + 1];

                hiddenHiddenGradientArray[i] = GetMatrix(rowCount, columnCount);
            }

            double[][] inputHiddenGradientArray = GetMatrix(this.inputNodeCount, this.hiddenNodeCountArray[0]);

            double[]   outputNodeBiasGradientArray = new double[this.outputNodeCount];
            double[][] hiddenNodeBiasGradientArray = GetJaggedMatrix(this.hiddenNodeCountArray);


            double[]   outputSignalArray = new double[this.outputNodeCount];
            double[][] hiddenSignalArray = GetJaggedMatrix(this.hiddenNodeCountArray);


            double[][]   hiddenOutputPreviousWeightDeltaArray = GetMatrix(this.hiddenNodeCountArray[this.hiddenLayerCount - 1], this.outputNodeCount);
            double[][][] hiddenHiddenPreviousWeightDeltaArray = new double[this.hiddenLayerCount - 1][][];

            for(int i = 0; i < hiddenHiddenPreviousWeightDeltaArray.Length; ++i)
            {
                int rowCount    = this.hiddenNodeCountArray[i];
                int columnCount = this.hiddenNodeCountArray[i + 1];

                hiddenHiddenPreviousWeightDeltaArray[i] = GetMatrix(rowCount, columnCount);
            }

            double[][] inputHiddenPreviousWeightDeltaArray = GetMatrix(this.inputNodeCount, this.hiddenNodeCountArray[0]);
            double[]   outputPreviousBiasDeltaArray        = new double[this.outputNodeCount];
            double[][] hiddenPreviousBiasDeltaArray        = GetJaggedMatrix(this.hiddenNodeCountArray);

            int epoch = 0;

            double[] inputValueArray  = new double[this.inputNodeCount ];
            double[] targetValueArray = new double[this.outputNodeCount];
            double   derivative       = 0d;
            double   errorSignal      = 0d;

            int[] sequenceArray       = new int[trainingDataArray.Length];
            int   sequenceArrayLength = sequenceArray.Length;

            for(int i = 0; i < sequenceArrayLength; ++i)
            {
                sequenceArray[i] = i;
            }

            int errorInterval = maximumEpochCount / showErrorCount;

            while(epoch < maximumEpochCount)
            {
                ++epoch;

                if(epoch % errorInterval == 0 && epoch < maximumEpochCount)
                {
                    double trainingError    = GetError(trainingDataArray, false);
                    double trainingAccuracy = GetAccuracy(trainingDataArray, false);

                    Console.WriteLine("반복 차수 = {0}, MS 오차 = {1:F4}, 정확도 = {2:F4}", epoch, trainingError, trainingAccuracy);
                    Console.WriteLine("입력-은닉 [0][0] 그라디언트 = {0:F6}", this.inputHiddenOneGradient);
                    Console.WriteLine();
                }

                Shuffle(sequenceArray);

                int trainingDataArrayLength = trainingDataArray.Length;

                for(int ii = 0; ii < trainingDataArrayLength; ++ii)
                {
                    int itemIndex = sequenceArray[ii];

                    Array.Copy(trainingDataArray[itemIndex], inputValueArray, this.inputNodeCount);

                    Array.Copy(trainingDataArray[itemIndex], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                    Query(inputValueArray);

                    for(int j = 0; j < this.outputNodeCount; ++j)
                    {
                        errorSignal = targetValueArray[j] - this.outputNodeArray[j];

                        derivative = (1 - this.outputNodeArray[j]) * this.outputNodeArray[j];

                        outputSignalArray[j] = errorSignal * derivative;
                    }

                    int lastLayer = this.hiddenLayerCount - 1;

                    for(int j = 0; j < this.hiddenNodeCountArray[lastLayer]; ++j)
                    {
                        derivative = (1 + this.hiddenNodeArray[lastLayer][j]) * (1 - this.hiddenNodeArray[lastLayer][j]);

                        double summary = 0d;

                        for(int k = 0; k < this.outputNodeCount; ++k)
                        {
                            summary += outputSignalArray[k] * this.hiddenOutputWeightArray[j][k];
                        }

                        hiddenSignalArray[lastLayer][j] = derivative * summary;
                    }

                    for(int j = lastLayer - 1; j >= 0; --j)
                    {
                        for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)
                        {
                            derivative = (1 + this.hiddenNodeArray[j][k]) * (1 - this.hiddenNodeArray[j][k]);

                            double summary = 0d;

                            for(int l = 0; l < this.hiddenNodeCountArray[j + 1]; ++l)
                            {
                                summary += hiddenSignalArray[j + 1][l] * this.hiddenHiddenWeightArray[j][k][l];
                            }

                            hiddenSignalArray[j][k] = derivative * summary;
                        }
                    }

                    for(int j = 0; j < this.inputNodeCount; ++j)
                    {
                        for(int k = 0; k < this.hiddenNodeCountArray[0]; ++k)
                        {
                            inputHiddenGradientArray[j][k] = this.inputNodeArray[j] * hiddenSignalArray[0][k];
                        }
                    }

                    this.inputHiddenOneGradient = inputHiddenGradientArray[0][0];

                    for(int j = 0; j < this.hiddenLayerCount - 1; ++j)
                    {
                        for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)
                        {
                            for(int l = 0; l < this.hiddenNodeCountArray[j + 1]; ++l)
                            {
                                hiddenHiddenGradientArray[j][k][l] = this.hiddenNodeArray[j][k] * hiddenSignalArray[j + 1][l];
                            }
                        }
                    }

                    for(int j = 0; j < this.hiddenNodeCountArray[lastLayer]; ++j)
                    {
                        for(int k = 0; k < this.outputNodeCount; ++k)
                        {
                            hiddenOutputGradientArray[j][k] = this.hiddenNodeArray[lastLayer][j] * outputSignalArray[k];
                        }
                    }

                    for(int j = 0; j < this.hiddenLayerCount; ++j)
                    {
                        for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)
                        {
                            hiddenNodeBiasGradientArray[j][k] = 1.0 * hiddenSignalArray[j][k];
                        }
                    }

                    for(int j = 0; j < this.outputNodeCount; ++j)
                    {
                        outputNodeBiasGradientArray[j] = 1.0 * outputSignalArray[j];
                    }

                    for(int j = 0; j < this.inputNodeCount; ++j)
                    {
                        for(int k = 0; k < this.hiddenNodeCountArray[0]; ++k)
                        {
                            double delta = inputHiddenGradientArray[j][k] * learningRate;

                            this.inputHiddenWeightArray[j][k] += delta;
                            this.inputHiddenWeightArray[j][k] += inputHiddenPreviousWeightDeltaArray[j][k] * momentum;

                            inputHiddenPreviousWeightDeltaArray[j][k] = delta;
                        }
                    }

                    for(int j = 0; j < this.hiddenLayerCount - 1; ++j)
                    {
                        for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)
                        {
                            for(int l = 0; l < this.hiddenNodeCountArray[j + 1]; ++l)
                            {
                                double delta = hiddenHiddenGradientArray[j][k][l] * learningRate;

                                this.hiddenHiddenWeightArray[j][k][l] += delta;
                                this.hiddenHiddenWeightArray[j][k][l] += hiddenHiddenPreviousWeightDeltaArray[j][k][l] * momentum;

                                hiddenHiddenPreviousWeightDeltaArray[j][k][l] = delta;
                            }
                        }
                    }

                    for(int j = 0; j < this.hiddenNodeCountArray[lastLayer]; ++j)
                    {
                        for(int k = 0; k < this.outputNodeCount; ++k)
                        {
                            double delta = hiddenOutputGradientArray[j][k] * learningRate;

                            this.hiddenOutputWeightArray[j][k] += delta;
                            this.hiddenOutputWeightArray[j][k] += hiddenOutputPreviousWeightDeltaArray[j][k] * momentum;

                            hiddenOutputPreviousWeightDeltaArray[j][k] = delta;
                        }
                    }

                    for(int j = 0; j < this.hiddenLayerCount; ++j)
                    {
                        for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)
                        {
                            double delta = hiddenNodeBiasGradientArray[j][k] * learningRate;

                            this.hiddenNodeBiasArray[j][k] += delta;
                            this.hiddenNodeBiasArray[j][k] += hiddenPreviousBiasDeltaArray[j][k] * momentum;

                            hiddenPreviousBiasDeltaArray[j][k] = delta;
                        }
                    }

                    for(int j = 0; j < this.outputNodeCount; ++j)
                    {
                        double delta = outputNodeBiasGradientArray[j] * learningRate;

                        this.outputNodeBiasArray[j] += delta;
                        this.outputNodeBiasArray[j] += outputPreviousBiasDeltaArray[j] * momentum;

                        outputPreviousBiasDeltaArray[j] = delta;
                    }
                }
            }

            double[] bestWeightArray = GetWeightArray();

            return bestWeightArray;
        }

        #endregion
        #region 덤프하기 - Dump()

        /// <summary>
        /// 덤프하기
        /// </summary>
        public void Dump()
        {
            for(int i = 0; i < this.inputNodeCount; ++i)
            {
                Console.WriteLine("입력 노드[{0}] = {1:F4}", i, this.inputNodeArray[i]);
            }

            for(int i = 0; i < this.hiddenLayerCount; ++i)
            {
                Console.WriteLine();

                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    Console.WriteLine("{0} 은닉층 노드[{1}] = {2:F4}", i, j, this.hiddenNodeArray[i][j]);
                }
            }

            Console.WriteLine();

            for(int i = 0; i < this.outputNodeCount; ++i)
            {
                Console.WriteLine("출력 노드[{0}] = {1:F4}", i, this.outputNodeArray[i]);
            }

            Console.WriteLine();

            for(int i = 0; i < this.inputNodeCount; ++i)
            {
                for(int j = 0; j < this.hiddenNodeCountArray[0]; ++j)
                {
                    Console.WriteLine("입력-은닉 가중치[{0}][{1}] = {2:F4}", i, j, this.inputHiddenWeightArray[i][j]);
                }
            }

            for(int i = 0; i < this.hiddenLayerCount - 1; ++i)
            {
                Console.WriteLine();

                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    for(int k = 0; k < this.hiddenNodeCountArray[i + 1]; ++k)
                    {
                        Console.WriteLine("은닉[{0}]-은닉[{1}]층 가중치 노드[{2}][{3}] = {4:F4}", i, (i + 1), j, k, this.hiddenHiddenWeightArray[i][j][k]);
                    }
                }
            }

            Console.WriteLine();

            for(int i = 0; i < this.hiddenNodeCountArray[this.hiddenLayerCount - 1]; ++i)
            {
                for(int j = 0; j < this.outputNodeCount; ++j)
                {
                    Console.WriteLine("은닉-출력 가중치[{0}][{1}] = {2:F4}", i, j, this.hiddenOutputWeightArray[i][j]);
                }
            }

            for(int i = 0; i < this.hiddenLayerCount; ++i)
            {
                Console.WriteLine();

                for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)
                {
                    Console.WriteLine("{0} 은닉층 바이어스[{1}] = {2:F4}", i, j, this.hiddenNodeBiasArray[i][j]);
                }
            }

            Console.WriteLine();

            for(int i = 0; i < this.outputNodeCount; ++i)
            {
                Console.WriteLine("출력 노드 바이어스[{0}] = {1:F4}", i, this.outputNodeBiasArray[i]);
            }
        }

        #endregion

        //////////////////////////////////////////////////////////////////////////////// Private

        #region 가중치 배열 초기화 하기 - InitializeWeightArray()

        /// <summary>
        /// 가중치 배열 초기화 하기
        /// </summary>
        private void InitializeWeightArray()
        {
            double low  = -0.10;
            double high = +0.10;

            int totalWeightCount = DeepNeuralNetwork.GetTotalWeightCount(this.inputNodeCount, this.hiddenNodeCountArray, this.outputNodeCount);

            double[] weightArray = new double[totalWeightCount];

            for(int i = 0; i < totalWeightCount; ++i)
            {
                weightArray[i] = (high - low) * _random.NextDouble() + low;
            }

            SetWeightArray(weightArray);
        }

        #endregion
        #region 섞기 - Shuffle(sequenceArray)

        /// <summary>
        /// 섞기
        /// </summary>
        /// <param name="sequenceArray">순차 배열</param>
        private void Shuffle(int[] sequenceArray)
        {
            int sequenceArrayLength = sequenceArray.Length;

            for(int i = 0; i < sequenceArrayLength; ++i)
            {
                int replaceIndex = _random.Next(i, sequenceArrayLength);
                int replaceValue = sequenceArray[replaceIndex];

                sequenceArray[replaceIndex] = sequenceArray[i];

                sequenceArray[i] = replaceValue;
            }
        }

        #endregion
    }
}

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789

790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

805

806

807

808

809

810

811

812

813

814

815

816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

832

833

834

835

836

837

838

839

840

841

842

843

844

845

846

847

848

849

850

851

852

853

854

855

856

857

858

859

860

861

862

863

864

865

866

867

868

869

870

871

872

873

874

875

876

877

878

879

880

881

882

883

884

885

886

887

888

889

890

891

892

893

894

895

896

897

898

899

900

901

902

903

904

905

906

907

908

909

910

911

912

913

914

915

916

917

918

919

920

921

922

923

924

925

926

927

928

929

930

931

932

933

934

935

936

937

938

939

940

941

942

943

944

945

946

947

948

949

950

951

952

953

954

955

956

957

958

959

960

961

962

963

964

965

966

967

968

969

970

971

972

973

974

975

976

977

978

979

980

981

982

983

984

985

986

987

988

989

990

991

992

993

994

995

996

997

998

999

1000

1001

1002

1003

1004

1005

1006

1007

1008

1009

1010

1011

1012

1013

1014

1015

1016

1017

1018

1019

1020

1021

1022

1023

1024

1025

1026

1027

1028

1029

1030

1031

1032

1033

1034

1035

1036

1037

1038

1039

1040

1041

1042

1043

1044

1045

1046

1047

1048

1049

1050

1051

1052

1053

1054

1055

1056

1057

1058

1059

1060

1061

1062

1063

1064

1065

1066

1067

1068

1069

1070

1071

1072

1073

1074

1075

1076

1077

1078

1079

1080

1081

1082

1083

1084

1085

1086

1087

1088

1089

1090

1091

1092

1093

1094

1095

1096

1097

1098

1099

1100

1101

1102

1103

1104

1105

1106

using System;

namespace TestProject

{

/// <summary>

/// 심층 신경망

/// </summary>

public class DeepNeuralNetwork

{

//////////////////////////////////////////////////////////////////////////////////////////////////// Field

////////////////////////////////////////////////////////////////////////////////////////// Static

//////////////////////////////////////////////////////////////////////////////// Private

#region Field

/// <summary>

/// 난수기

/// </summary>

private static Random _random;

#endregion

////////////////////////////////////////////////////////////////////////////////////////// Instance

//////////////////////////////////////////////////////////////////////////////// Private

#region Field

/// <summary>

/// 입력 노드 수

/// </summary>

private int inputNodeCount;

/// <summary>

/// 은닉 노드 수 배열

/// </summary>

private int[] hiddenNodeCountArray;

/// <summary>

/// 출력 노드 수

/// </summary>

private int outputNodeCount;

/// <summary>

/// 은닉층 수

/// </summary>

private int hiddenLayerCount;

/// <summary>

/// 입력 노드 배열

/// </summary>

private double[] inputNodeArray;

/// <summary>

/// 은닉 노드 배열

/// </summary>

private double[][] hiddenNodeArray;

/// <summary>

/// 출력 노드 배열

/// </summary>

private double[] outputNodeArray;

/// <summary>

/// 입력/은닉 가중치 배열

/// </summary>

private double[][] inputHiddenWeightArray;

/// <summary>

/// 은닉/은닉 가중치 배열

/// </summary>

private double[][][] hiddenHiddenWeightArray;

/// <summary>

/// 은닉/출력 가중치 배열

/// </summary>

private double[][] hiddenOutputWeightArray;

/// <summary>

/// 은닉 노드 바이어스 배열

/// </summary>

private double[][] hiddenNodeBiasArray;

/// <summary>

/// 출력 노드 바이어스 배열

/// </summary>

private double[] outputNodeBiasArray;

/// <summary>

/// 입력/은닉 1개 그라디언트

/// </summary>

private double inputHiddenOneGradient;

#endregion

//////////////////////////////////////////////////////////////////////////////////////////////////// Constructor

////////////////////////////////////////////////////////////////////////////////////////// Public

#region 생성자 - DeepNeuralNetwork(inputNodeCount, hiddenNodeCountArray, outputNodeCount)

/// <summary>

/// 생성자

/// </summary>

/// <param name="inputNodeCount">입력 노드 수</param>

/// <param name="hiddenNodeCountArray">은닉 노드 수 배열</param>

/// <param name="outputNodeCount">출력 노드 수</param>

public DeepNeuralNetwork(int inputNodeCount, int[] hiddenNodeCountArray, int outputNodeCount)

{

_random = new Random(0);

this.inputNodeCount = inputNodeCount;

this.hiddenNodeCountArray = new int[hiddenNodeCountArray.Length];

for(int i = 0; i < hiddenNodeCountArray.Length; ++i)

{

this.hiddenNodeCountArray[i] = hiddenNodeCountArray[i];

}

this.outputNodeCount = outputNodeCount;

this.hiddenLayerCount = hiddenNodeCountArray.Length;

this.inputNodeArray = new double[inputNodeCount];

this.hiddenNodeArray = GetJaggedMatrix(hiddenNodeCountArray);

this.outputNodeArray = new double[outputNodeCount];

this.inputHiddenWeightArray = GetMatrix(inputNodeCount, hiddenNodeCountArray[0]);

this.hiddenOutputWeightArray = GetMatrix(hiddenNodeCountArray[this.hiddenLayerCount - 1], outputNodeCount);

this.hiddenHiddenWeightArray = new double[this.hiddenLayerCount - 1][][];

for(int i = 0; i < this.hiddenHiddenWeightArray.Length; ++i)

{

int rowCount = hiddenNodeCountArray[i ];

int columnCount = hiddenNodeCountArray[i + 1];

this.hiddenHiddenWeightArray[i] = GetMatrix(rowCount, columnCount);

}

this.hiddenNodeBiasArray = GetJaggedMatrix(hiddenNodeCountArray);

this.outputNodeBiasArray = new double[outputNodeCount];

InitializeWeightArray();

}

#endregion

//////////////////////////////////////////////////////////////////////////////////////////////////// Method

////////////////////////////////////////////////////////////////////////////////////////// Static

//////////////////////////////////////////////////////////////////////////////// Public

#region 전체 가중치 수 구하기 - GetTotalWeightCount(inputNodeCount, hiddenNodeCountArray, outputNodeCount)

/// <summary>

/// 전체 가중치 수 구하기

/// </summary>

/// <param name="inputNodeCount">입력 노드 수</param>

/// <param name="hiddenNodeCountArray">은닉 노드 수 배열</param>

/// <param name="outputNodeCount">출력 노드 수</param>

/// <returns>전체 가중치 수</returns>

public static int GetTotalWeightCount(int inputNodeCount, int[] hiddenNodeCountArray, int outputNodeCount)

{

int inputHeightWeightCount = inputNodeCount * hiddenNodeCountArray[0];

int hiddenNodeCountArrayLength = hiddenNodeCountArray.Length;

int hiddenHiddenWeightCount = 0;

for(int i = 0; i < hiddenNodeCountArrayLength - 1; ++i)

{

int rowCount = hiddenNodeCountArray[i ];

int columnCount = hiddenNodeCountArray[i + 1];

hiddenHiddenWeightCount += rowCount * columnCount;

}

int hiddenOutputWeightCount = hiddenNodeCountArray[hiddenNodeCountArrayLength - 1] * outputNodeCount;

int hiddenBiasCount = 0;

for(int i = 0; i < hiddenNodeCountArrayLength; ++i)

{

hiddenBiasCount += hiddenNodeCountArray[i];

}

int outputBiasCount = outputNodeCount;

int totalWeightCount = inputHeightWeightCount + hiddenHiddenWeightCount + hiddenOutputWeightCount + hiddenBiasCount + outputBiasCount;

return totalWeightCount;

}

#endregion

//////////////////////////////////////////////////////////////////////////////// Private

#region 배열 표시하기 - DisplayArray(sourceArray, decimalPointCount)

/// <summary>

/// 배열 표시하기

/// </summary>

/// <param name="sourceArray">소스 배열</param>

/// <param name="decimalPointCount">소수점 자리수</param>

private static void DisplayArray(double[] sourceArray, int decimalPointCount)

{

string displayFormat = string.Format("F{0}", decimalPointCount);

for(int i = 0; i < sourceArray.Length; ++i)

{

Console.Write(sourceArray[i].ToString(displayFormat) + " ");

}

Console.WriteLine();

}

#endregion

#region JAGGED 매트릭스 구하기 - GetJaggedMatrix(columnCountArray)

/// <summary>

/// JAGGED 매트릭스 구하기

/// </summary>

/// <param name="columnCountArray">컬럼 수 배열</param>

/// <returns>JAGGED 매트릭스</returns>

private static double[][] GetJaggedMatrix(int[] columnCountArray)

{

int rowCount = columnCountArray.Length;

double[][] resultArray = new double[rowCount][];

for(int i = 0; i < rowCount; ++i)

{

int columnCount = columnCountArray[i];

resultArray[i] = new double[columnCount];

}

return resultArray;

}

#endregion

#region 매트릭스 구하기 - GetMatrix(rowCount, columnCount)

/// <summary>

/// 매트릭스 구하기

/// </summary>

/// <param name="rowCount">행 수</param>

/// <param name="columnCount">컬럼 수</param>

/// <returns>매트릭스</returns>

private static double[][] GetMatrix(int rowCount, int columnCount)

{

double[][] resultArray = new double[rowCount][];

for(int i = 0; i < rowCount; ++i)

{

resultArray[i] = new double[columnCount];

}

return resultArray;

}

#endregion

#region 최대 값 인덱스 구하기 - GetMaximumValueIndex(sourceArray)

/// <summary>

/// 최대 값 인덱스 구하기

/// </summary>

/// <param name="sourceArray">소스 배열</param>

/// <returns>최대 값 인덱스</returns>

private static int GetMaximumValueIndex(double[] sourceArray)

{

int maximumValueIndex = 0;

double maximumValue = sourceArray[0];

int sourceArrayLength = sourceArray.Length;

for(int i = 0; i < sourceArrayLength; ++i)

{

if(sourceArray[i] > maximumValue)

{

maximumValue = sourceArray[i];

maximumValueIndex = i;

}

return maximumValueIndex;

}

#endregion

#region 하이퍼볼릭 탄젠트 구하기 - GetHyperbolicTangent(x)

/// <summary>

/// 하이퍼볼릭 탄젠트 구하기

/// </summary>

/// <param name="x">X</param>

/// <returns>하이퍼볼릭 탄젠트</returns>

private static double GetHyperbolicTangent(double x)

{

if(x < -20d)

{

return -1d;

}

else if(x > 20d)

{

return 1d;

}

else

{

return Math.Tanh(x);

}

#endregion

#region 소프트맥스 구하기 - GetSoftmax(sourceArray)

/// <summary>

/// 소프트맥스 구하기

/// </summary>

/// <param name="sourceArray">소스 배열</param>

/// <returns>소프트맥스</returns>

private static double[] GetSoftmax(double[] sourceArray)

{

int sourceArrayLength = sourceArray.Length;

double summary = 0d;

for(int i = 0; i < sourceArrayLength; ++i)

{

summary += Math.Exp(sourceArray[i]);

}

double[] resultArray = new double[sourceArrayLength];

for(int i = 0; i < sourceArrayLength; ++i)

{

resultArray[i] = Math.Exp(sourceArray[i]) / summary;

}

return resultArray;

}

#endregion

////////////////////////////////////////////////////////////////////////////////////////// Instance

//////////////////////////////////////////////////////////////////////////////// Public

#region 가중치 배열 설정하기 - SetWeightArray(sourceArray)

/// <summary>

/// 가중치 배열 설정하기

/// </summary>

/// <param name="sourceArray">소스 배열</param>

public void SetWeightArray(double[] sourceArray)

{

int totalWeightCount = GetTotalWeightCount(this.inputNodeCount, this.hiddenNodeCountArray, this.outputNodeCount);

if(sourceArray.Length != totalWeightCount)

{

throw new Exception("부적절한 가중치 배열 길이 입니다.");

}

int firstHiddenNodeArrayLength = this.hiddenNodeArray[0].Length;

int weightArrayIndex = 0;

for(int i = 0; i < this.inputNodeCount; ++i)

{

for(int j = 0; j < firstHiddenNodeArrayLength; ++j)

{

this.inputHiddenWeightArray[i][j] = sourceArray[weightArrayIndex++];

}

for(int i = 0; i < this.hiddenLayerCount - 1; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[i + 1]; ++k)

{

this.hiddenHiddenWeightArray[i][j][k] = sourceArray[weightArrayIndex++];

}

int hiddenNodeCountArrayLastIndex = this.hiddenLayerCount - 1;

for(int i = 0; i < this.hiddenNodeCountArray[hiddenNodeCountArrayLastIndex]; ++i)

{

for(int j = 0; j < this.outputNodeCount; ++j)

{

this.hiddenOutputWeightArray[i][j] = sourceArray[weightArrayIndex++];

}

for(int i = 0; i < this.hiddenLayerCount; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

this.hiddenNodeBiasArray[i][j] = sourceArray[weightArrayIndex++];

}

for(int i = 0; i < this.outputNodeCount; ++i)

{

this.outputNodeBiasArray[i] = sourceArray[weightArrayIndex++];

}

#endregion

#region 질의하기 - Query(inputValueArray)

/// <summary>

/// 질의하기

/// </summary>

/// <param name="inputValueArray">입력 값 배열</param>

/// <returns>출력 값 배열</returns>

public double[] Query(double[] inputValueArray)

{

for(int i = 0; i < this.inputNodeCount; ++i)

{

this.inputNodeArray[i] = inputValueArray[i];

}

for(int i = 0; i < this.hiddenLayerCount; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

this.hiddenNodeArray[i][j] = 0d;

}

for(int i = 0; i < this.outputNodeCount; ++i)

{

this.outputNodeArray[i] = 0.0;

}

for(int i = 0; i < this.hiddenNodeCountArray[0]; ++i)

{

for(int j = 0; j < this.inputNodeCount; ++j)

{

this.hiddenNodeArray[0][i] += this.inputHiddenWeightArray[j][i] * this.inputNodeArray[j];

}

this.hiddenNodeArray[0][i] += this.hiddenNodeBiasArray[0][i];

this.hiddenNodeArray[0][i] = Math.Tanh(this.hiddenNodeArray[0][i]);

}

for(int i = 1; i < this.hiddenLayerCount; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[i - 1]; ++k)

{

this.hiddenNodeArray[i][j] += this.hiddenHiddenWeightArray[i - 1][k][j] * this.hiddenNodeArray[i - 1][k];

}

this.hiddenNodeArray[i][j] += this.hiddenNodeBiasArray[i][j];

this.hiddenNodeArray[i][j] = Math.Tanh(this.hiddenNodeArray[i][j]);

}

for(int i = 0; i < this.outputNodeCount; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[this.hiddenLayerCount - 1]; ++j)

{

this.outputNodeArray[i] += this.hiddenOutputWeightArray[j][i] * this.hiddenNodeArray[this.hiddenLayerCount - 1][j];

}

this.outputNodeArray[i] += this.outputNodeBiasArray[i];

}

double[] outputArray = GetSoftmax(this.outputNodeArray);

for(int k = 0; k < this.outputNodeCount; ++k)

{

this.outputNodeArray[k] = outputArray[k];

}

return outputArray;

}

#endregion

#region 오차 구하기 - GetError(trainingDataArray, verbose)

/// <summary>

/// 오차 구하기

/// </summary>

/// <param name="trainingDataArray">훈련 데이터 배열</param>

/// <param name="verbose">서술 여부</param>

/// <returns>오차</returns>

public double GetError(double[][] trainingDataArray, bool verbose)

{

double summarySquaredError = 0d;

double[] inputValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

for(int i = 0; i < trainingDataArray.Length; ++i)

{

Array.Copy(trainingDataArray[i], inputValueArray, this.inputNodeCount);

Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

double[] outputValueArray = this.Query(inputValueArray);

if(verbose == true)

{

DisplayArray(outputValueArray, 4);

DisplayArray(targetValueArray, 4);

Console.WriteLine();

Console.ReadLine();

}

for(int j = 0; j < this.outputNodeCount; ++j)

{

double error = targetValueArray[j] - outputValueArray[j];

summarySquaredError += error * error;

}

return summarySquaredError / (trainingDataArray.Length * this.outputNodeCount);

}

#endregion

#region 오차 구하기 - GetError(trainingDataArray, weightArray)

/// <summary>

/// 오차 구하기

/// </summary>

/// <param name="trainingDataArray">훈련 데이터 배열</param>

/// <param name="weightArray">가중치 배열</param>

/// <returns>오차</returns>

public double GetError(double[][] trainingDataArray, double[] weightArray)

{

SetWeightArray(weightArray);

double summarySquaredError = 0d;

double[] intputValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

int trainingDataArrayLength = trainingDataArray.Length;

for(int i = 0; i < trainingDataArrayLength; ++i)

{

Array.Copy(trainingDataArray[i], intputValueArray, this.inputNodeCount);

Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

double[] outputValueArray = Query(intputValueArray);

for(int j = 0; j < this.outputNodeCount; ++j)

{

double error = targetValueArray[j] - outputValueArray[j];

summarySquaredError += error * error;

}

return summarySquaredError / trainingDataArrayLength;

}

#endregion

#region 정확도 구하기 - GetAccuracy(trainingDataArray, verbose)

/// <summary>

/// 정확도 구하기

/// </summary>

/// <param name="trainingDataArray">훈련 데이터 배열</param>

/// <param name="verbose">서술 여부</param>

/// <returns>정확도</returns>

public double GetAccuracy(double[][] trainingDataArray, bool verbose)

{

int correctCount = 0;

int wrongCount = 0;

double[] intputValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

double[] outputValueArray;

int trainingDataArrayLength = trainingDataArray.Length;

for(int i = 0; i < trainingDataArrayLength; ++i)

{

Array.Copy(trainingDataArray[i], intputValueArray, this.inputNodeCount);

Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

outputValueArray = Query(intputValueArray);

if(verbose == true)

{

DisplayArray(outputValueArray, 4);

DisplayArray(targetValueArray, 4);

Console.WriteLine();

Console.ReadLine();

}

int outputMaximumValueIndex = GetMaximumValueIndex(outputValueArray);

int targetMaximumValueIndex = GetMaximumValueIndex(targetValueArray);

if(outputMaximumValueIndex == targetMaximumValueIndex)

{

++correctCount;

}

else

{

++wrongCount;

}

return (correctCount * 1d) / (correctCount + wrongCount);

}

#endregion

#region 정확도 구하기 - GetAccuracy(trainingDataArray, weightArray)

/// <summary>

/// 정확도 구하기

/// </summary>

/// <param name="trainingDataArray">훈련 데이터 배열</param>

/// <param name="weightArray">가중치 배열</param>

/// <returns>정확도</returns>

public double GetAccuracy(double[][] trainingDataArray, double[] weightArray)

{

SetWeightArray(weightArray);

int correctCount = 0;

int wrongCount = 0;

double[] inputValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

double[] outputValueArray;

int trainingDataArrayLength = trainingDataArray.Length;

for(int i = 0; i < trainingDataArrayLength; ++i)

{

Array.Copy(trainingDataArray[i], inputValueArray, this.inputNodeCount);

Array.Copy(trainingDataArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

outputValueArray = Query(inputValueArray);

int outputMaximumValueIndex = GetMaximumValueIndex(outputValueArray);

int targetMaximumValueIndex = GetMaximumValueIndex(targetValueArray);

if(outputMaximumValueIndex == targetMaximumValueIndex)

{

++correctCount;

}

else

{

++wrongCount;

}

return (correctCount * 1d) / (correctCount + wrongCount);

}

#endregion

#region 가중치 배열 구하기 - GetWeightArray()

/// <summary>

/// 가중치 배열 구하기

/// </summary>

/// <returns>가중치 배열</returns>

public double[] GetWeightArray()

{

int totalWeightCount = GetTotalWeightCount(this.inputNodeCount, this.hiddenNodeCountArray, this.outputNodeCount);

double[] resultArray = new double[totalWeightCount];

int index = 0;

for(int i = 0; i < this.inputNodeCount; ++i)

{

for(int j = 0; j < this.hiddenNodeArray[0].Length; ++j)

{

resultArray[index++] = this.inputHiddenWeightArray[i][j];

}

for(int i = 0; i < this.hiddenLayerCount - 1; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[i + 1]; ++k)

{

resultArray[index++] = this.hiddenHiddenWeightArray[i][j][k];

}

int hiddenNodeCountArrayLastIndex = this.hiddenLayerCount - 1;

for(int i = 0; i < this.hiddenNodeCountArray[hiddenNodeCountArrayLastIndex]; ++i)

{

for(int j = 0; j < this.outputNodeCount; ++j)

{

resultArray[index++] = this.hiddenOutputWeightArray[i][j];

}

for(int i = 0; i < this.hiddenLayerCount; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

resultArray[index++] = this.hiddenNodeBiasArray[i][j];

}

for(int i = 0; i < this.outputNodeCount; ++i)

{

resultArray[index++] = this.outputNodeBiasArray[i];

}

return resultArray;

}

#endregion

#region 훈련하기 - Train(trainingDataArray, maximumEpochCount, learningRate, momentum, showErrorCount)

/// <summary>

/// 훈련하기

/// </summary>

/// <param name="trainingDataArray">훈련 데이터 배열</param>

/// <param name="maximumEpochCount">최대 반복 차수</param>

/// <param name="learningRate">학습률</param>

/// <param name="momentum">모멘텀</param>

/// <param name="showErrorCount">오차 표시 수</param>

/// <returns>최적 가중치 배열</returns>

public double[] Train(double[][] trainingDataArray, int maximumEpochCount, double learningRate, double momentum, int showErrorCount)

{

double[][] hiddenOutputGradientArray = GetMatrix(this.hiddenNodeCountArray[this.hiddenLayerCount - 1], this.outputNodeCount);

double[][][] hiddenHiddenGradientArray = new double[this.hiddenLayerCount - 1][][];

for(int i = 0; i < hiddenHiddenGradientArray.Length; ++i)

{

int rowCount = this.hiddenNodeCountArray[i ];

int columnCount = this.hiddenNodeCountArray[i + 1];

hiddenHiddenGradientArray[i] = GetMatrix(rowCount, columnCount);

}

double[][] inputHiddenGradientArray = GetMatrix(this.inputNodeCount, this.hiddenNodeCountArray[0]);

double[] outputNodeBiasGradientArray = new double[this.outputNodeCount];

double[][] hiddenNodeBiasGradientArray = GetJaggedMatrix(this.hiddenNodeCountArray);

double[] outputSignalArray = new double[this.outputNodeCount];

double[][] hiddenSignalArray = GetJaggedMatrix(this.hiddenNodeCountArray);

double[][] hiddenOutputPreviousWeightDeltaArray = GetMatrix(this.hiddenNodeCountArray[this.hiddenLayerCount - 1], this.outputNodeCount);

double[][][] hiddenHiddenPreviousWeightDeltaArray = new double[this.hiddenLayerCount - 1][][];

for(int i = 0; i < hiddenHiddenPreviousWeightDeltaArray.Length; ++i)

{

int rowCount = this.hiddenNodeCountArray[i];

int columnCount = this.hiddenNodeCountArray[i + 1];

hiddenHiddenPreviousWeightDeltaArray[i] = GetMatrix(rowCount, columnCount);

}

double[][] inputHiddenPreviousWeightDeltaArray = GetMatrix(this.inputNodeCount, this.hiddenNodeCountArray[0]);

double[] outputPreviousBiasDeltaArray = new double[this.outputNodeCount];

double[][] hiddenPreviousBiasDeltaArray = GetJaggedMatrix(this.hiddenNodeCountArray);

int epoch = 0;

double[] inputValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

double derivative = 0d;

double errorSignal = 0d;

int[] sequenceArray = new int[trainingDataArray.Length];

int sequenceArrayLength = sequenceArray.Length;

for(int i = 0; i < sequenceArrayLength; ++i)

{

sequenceArray[i] = i;

}

int errorInterval = maximumEpochCount / showErrorCount;

while(epoch < maximumEpochCount)

{

++epoch;

if(epoch % errorInterval == 0 && epoch < maximumEpochCount)

{

double trainingError = GetError(trainingDataArray, false);

double trainingAccuracy = GetAccuracy(trainingDataArray, false);

Console.WriteLine("반복 차수 = {0}, MS 오차 = {1:F4}, 정확도 = {2:F4}", epoch, trainingError, trainingAccuracy);

Console.WriteLine("입력-은닉 [0][0] 그라디언트 = {0:F6}", this.inputHiddenOneGradient);

Console.WriteLine();

}

Shuffle(sequenceArray);

int trainingDataArrayLength = trainingDataArray.Length;

for(int ii = 0; ii < trainingDataArrayLength; ++ii)

{

int itemIndex = sequenceArray[ii];

Array.Copy(trainingDataArray[itemIndex], inputValueArray, this.inputNodeCount);

Array.Copy(trainingDataArray[itemIndex], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

Query(inputValueArray);

for(int j = 0; j < this.outputNodeCount; ++j)

{

errorSignal = targetValueArray[j] - this.outputNodeArray[j];

derivative = (1 - this.outputNodeArray[j]) * this.outputNodeArray[j];

outputSignalArray[j] = errorSignal * derivative;

}

int lastLayer = this.hiddenLayerCount - 1;

for(int j = 0; j < this.hiddenNodeCountArray[lastLayer]; ++j)

{

derivative = (1 + this.hiddenNodeArray[lastLayer][j]) * (1 - this.hiddenNodeArray[lastLayer][j]);

double summary = 0d;

for(int k = 0; k < this.outputNodeCount; ++k)

{

summary += outputSignalArray[k] * this.hiddenOutputWeightArray[j][k];

}

hiddenSignalArray[lastLayer][j] = derivative * summary;

}

for(int j = lastLayer - 1; j >= 0; --j)

{

for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)

{

derivative = (1 + this.hiddenNodeArray[j][k]) * (1 - this.hiddenNodeArray[j][k]);

double summary = 0d;

for(int l = 0; l < this.hiddenNodeCountArray[j + 1]; ++l)

{

summary += hiddenSignalArray[j + 1][l] * this.hiddenHiddenWeightArray[j][k][l];

}

hiddenSignalArray[j][k] = derivative * summary;

}

for(int j = 0; j < this.inputNodeCount; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[0]; ++k)

{

inputHiddenGradientArray[j][k] = this.inputNodeArray[j] * hiddenSignalArray[0][k];

}

this.inputHiddenOneGradient = inputHiddenGradientArray[0][0];

for(int j = 0; j < this.hiddenLayerCount - 1; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)

{

for(int l = 0; l < this.hiddenNodeCountArray[j + 1]; ++l)

{

hiddenHiddenGradientArray[j][k][l] = this.hiddenNodeArray[j][k] * hiddenSignalArray[j + 1][l];

}

for(int j = 0; j < this.hiddenNodeCountArray[lastLayer]; ++j)

{

for(int k = 0; k < this.outputNodeCount; ++k)

{

hiddenOutputGradientArray[j][k] = this.hiddenNodeArray[lastLayer][j] * outputSignalArray[k];

}

for(int j = 0; j < this.hiddenLayerCount; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)

{

hiddenNodeBiasGradientArray[j][k] = 1.0 * hiddenSignalArray[j][k];

}

for(int j = 0; j < this.outputNodeCount; ++j)

{

outputNodeBiasGradientArray[j] = 1.0 * outputSignalArray[j];

}

for(int j = 0; j < this.inputNodeCount; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[0]; ++k)

{

double delta = inputHiddenGradientArray[j][k] * learningRate;

this.inputHiddenWeightArray[j][k] += delta;

this.inputHiddenWeightArray[j][k] += inputHiddenPreviousWeightDeltaArray[j][k] * momentum;

inputHiddenPreviousWeightDeltaArray[j][k] = delta;

}

for(int j = 0; j < this.hiddenLayerCount - 1; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)

{

for(int l = 0; l < this.hiddenNodeCountArray[j + 1]; ++l)

{

double delta = hiddenHiddenGradientArray[j][k][l] * learningRate;

this.hiddenHiddenWeightArray[j][k][l] += delta;

this.hiddenHiddenWeightArray[j][k][l] += hiddenHiddenPreviousWeightDeltaArray[j][k][l] * momentum;

hiddenHiddenPreviousWeightDeltaArray[j][k][l] = delta;

}

for(int j = 0; j < this.hiddenNodeCountArray[lastLayer]; ++j)

{

for(int k = 0; k < this.outputNodeCount; ++k)

{

double delta = hiddenOutputGradientArray[j][k] * learningRate;

this.hiddenOutputWeightArray[j][k] += delta;

this.hiddenOutputWeightArray[j][k] += hiddenOutputPreviousWeightDeltaArray[j][k] * momentum;

hiddenOutputPreviousWeightDeltaArray[j][k] = delta;

}

for(int j = 0; j < this.hiddenLayerCount; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[j]; ++k)

{

double delta = hiddenNodeBiasGradientArray[j][k] * learningRate;

this.hiddenNodeBiasArray[j][k] += delta;

this.hiddenNodeBiasArray[j][k] += hiddenPreviousBiasDeltaArray[j][k] * momentum;

hiddenPreviousBiasDeltaArray[j][k] = delta;

}

for(int j = 0; j < this.outputNodeCount; ++j)

{

double delta = outputNodeBiasGradientArray[j] * learningRate;

this.outputNodeBiasArray[j] += delta;

this.outputNodeBiasArray[j] += outputPreviousBiasDeltaArray[j] * momentum;

outputPreviousBiasDeltaArray[j] = delta;

}

double[] bestWeightArray = GetWeightArray();

return bestWeightArray;

}

#endregion

#region 덤프하기 - Dump()

/// <summary>

/// 덤프하기

/// </summary>

public void Dump()

{

for(int i = 0; i < this.inputNodeCount; ++i)

{

Console.WriteLine("입력 노드[{0}] = {1:F4}", i, this.inputNodeArray[i]);

}

for(int i = 0; i < this.hiddenLayerCount; ++i)

{

Console.WriteLine();

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

Console.WriteLine("{0} 은닉층 노드[{1}] = {2:F4}", i, j, this.hiddenNodeArray[i][j]);

}

Console.WriteLine();

for(int i = 0; i < this.outputNodeCount; ++i)

{

Console.WriteLine("출력 노드[{0}] = {1:F4}", i, this.outputNodeArray[i]);

}

Console.WriteLine();

for(int i = 0; i < this.inputNodeCount; ++i)

{

for(int j = 0; j < this.hiddenNodeCountArray[0]; ++j)

{

Console.WriteLine("입력-은닉 가중치[{0}][{1}] = {2:F4}", i, j, this.inputHiddenWeightArray[i][j]);

}

for(int i = 0; i < this.hiddenLayerCount - 1; ++i)

{

Console.WriteLine();

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

for(int k = 0; k < this.hiddenNodeCountArray[i + 1]; ++k)

{

Console.WriteLine("은닉[{0}]-은닉[{1}]층 가중치 노드[{2}][{3}] = {4:F4}", i, (i + 1), j, k, this.hiddenHiddenWeightArray[i][j][k]);

}

Console.WriteLine();

for(int i = 0; i < this.hiddenNodeCountArray[this.hiddenLayerCount - 1]; ++i)

{

for(int j = 0; j < this.outputNodeCount; ++j)

{

Console.WriteLine("은닉-출력 가중치[{0}][{1}] = {2:F4}", i, j, this.hiddenOutputWeightArray[i][j]);

}

for(int i = 0; i < this.hiddenLayerCount; ++i)

{

Console.WriteLine();

for(int j = 0; j < this.hiddenNodeCountArray[i]; ++j)

{

Console.WriteLine("{0} 은닉층 바이어스[{1}] = {2:F4}", i, j, this.hiddenNodeBiasArray[i][j]);

}

Console.WriteLine();

for(int i = 0; i < this.outputNodeCount; ++i)

{

Console.WriteLine("출력 노드 바이어스[{0}] = {1:F4}", i, this.outputNodeBiasArray[i]);

}

#endregion

//////////////////////////////////////////////////////////////////////////////// Private

#region 가중치 배열 초기화 하기 - InitializeWeightArray()

/// <summary>

/// 가중치 배열 초기화 하기

/// </summary>

private void InitializeWeightArray()

{

double low = -0.10;

double high = +0.10;

int totalWeightCount = DeepNeuralNetwork.GetTotalWeightCount(this.inputNodeCount, this.hiddenNodeCountArray, this.outputNodeCount);

double[] weightArray = new double[totalWeightCount];

for(int i = 0; i < totalWeightCount; ++i)

{

weightArray[i] = (high - low) * _random.NextDouble() + low;

}

SetWeightArray(weightArray);

}

#endregion

#region 섞기 - Shuffle(sequenceArray)

/// <summary>

/// 섞기

/// </summary>

/// <param name="sequenceArray">순차 배열</param>

private void Shuffle(int[] sequenceArray)

{

int sequenceArrayLength = sequenceArray.Length;

for(int i = 0; i < sequenceArrayLength; ++i)

{

int replaceIndex = _random.Next(i, sequenceArrayLength);

int replaceValue = sequenceArray[replaceIndex];

sequenceArray[replaceIndex] = sequenceArray[i];

sequenceArray[i] = replaceValue;

}

#endregion

}

▶ Program.cs


using System;

namespace TestProject
{
    /// <summary>
    /// 프로그램
    /// </summary>
    class Program
    {
        //////////////////////////////////////////////////////////////////////////////////////////////////// Method
        ////////////////////////////////////////////////////////////////////////////////////////// Static
        //////////////////////////////////////////////////////////////////////////////// Private

        #region 프로그램 시작하기 - Main()

        /// <summary>
        /// 프로그램 시작하기
        /// </summary>
        private static void Main()
        {
            Console.WriteLine();
            Console.WriteLine("심층 신경망(Deep Neural Network) 예제를 시작한다.");
            Console.WriteLine();

            int   inputNodeCount       = 4;
            int[] hiddenNodeCountArray = new int[] { 10, 10, 10 };
            int   outputNodeCount      = 3;

            int dataItemCount = 2000;

            Console.WriteLine("인공지능 훈련 데이터 항목 생성 시작 : {0}건", dataItemCount);

            double[][] trainingDataArray = GetTrainingDataArray(dataItemCount, inputNodeCount, hiddenNodeCountArray, outputNodeCount, 5);

            Console.WriteLine();
            Console.WriteLine("인공지능 훈련 데이터 항목 생성 완료");
            Console.WriteLine();
            Console.WriteLine("인공지능 훈련 데이터 항목 : ");
            Console.WriteLine();

            DisplayTrainingDataArray(trainingDataArray, 3, 2, true);

            Console.WriteLine();
            Console.WriteLine("4-(10,10,10)-3 심층 신경망을 생성한다.");
            Console.WriteLine();

            DeepNeuralNetwork dnn = new DeepNeuralNetwork(inputNodeCount, hiddenNodeCountArray, outputNodeCount);

            int    maximumEpochCount = 2000;
            double learningRate      = 0.001;
            double momentum          = 0.01;

            Console.WriteLine("최대 반복 차수 : {0}"   , maximumEpochCount);
            Console.WriteLine("학습률         : {0:F3}", learningRate     );
            Console.WriteLine("모멘텀         : {0:F3}", momentum);

            Console.WriteLine();
            Console.WriteLine("평균 제곱 오차를 갖는 역전파를 사용해 훈련을 시작한다.");
            Console.WriteLine();

            double[] bestWeightArray = dnn.Train(trainingDataArray, maximumEpochCount, learningRate, momentum, 10);

            Console.WriteLine("훈련 완료.");
            Console.WriteLine();

            double trainingError    = dnn.GetError(trainingDataArray, false);
            double trainingAccuracy = dnn.GetAccuracy(trainingDataArray, false);

            Console.WriteLine("최종 모델 MS 오차 : {0:F4}", trainingError   );
            Console.WriteLine("최종 모델 정확도  : {0:F4}", trainingAccuracy);

            Console.ReadLine();
        }

        #endregion

        #region 최대 값 인덱스 구하기 - GetMaximumValueIndex(sourceArray)

        /// <summary>
        /// 최대 값 인덱스 구하기
        /// </summary>
        /// <param name="sourceArray">소스 배열</param>
        /// <returns>최대 값 인덱스</returns>
        private static int GetMaximumValueIndex(double[] sourceArray)
        {
            int    sourceArrayLength = sourceArray.Length;
            int    maximumValueIndex = 0;
            double maximumValue      = sourceArray[0];

            for(int i = 0; i < sourceArrayLength; ++i)
            {
                if(sourceArray[i] > maximumValue)
                {
                    maximumValue = sourceArray[i];

                    maximumValueIndex = i;
                }
            }

            return maximumValueIndex;
        }

        #endregion
        #region 결과 값 배열 변환하기 - ConvertOutputValueArray(sourceArray)

        /// <summary>
        /// 결과 값 배열 변환하기
        /// </summary>
        /// <param name="sourceArray">소스 배열</param>
        /// <returns>변환 결과 값 배열</returns>
        private static double[] ConvertOutputValueArray(double[] sourceArray)
        {
            double[] resultArray = new double[sourceArray.Length];

            int maximumValueIndex = GetMaximumValueIndex(sourceArray);

            resultArray[maximumValueIndex] = 1d;

            return resultArray;
        }

        #endregion
        #region 훈련 데이터 배열 구하기 - GetTrainingDataArray(itemCount, inputNodeCount, hiddenNodeCountArray, outputNodeCount, seed)

        /// <summary>
        /// 훈련 데이터 배열 구하기
        /// </summary>
        /// <param name="itemCount">항목 수</param>
        /// <param name="inputNodeCount">입력 노드 수</param>
        /// <param name="hiddenNodeCountArray">은닉 노드 수 배열</param>
        /// <param name="outputNodeCount">출력 노드 수</param>
        /// <param name="seed">시드</param>
        /// <returns>훈련 데이터 배열</returns>
        private static double[][] GetTrainingDataArray(int itemCount, int inputNodeCount, int[] hiddenNodeCountArray, int outputNodeCount, int seed)
        {
            DeepNeuralNetwork dnn = new DeepNeuralNetwork(inputNodeCount, hiddenNodeCountArray, outputNodeCount);

            Random random = new Random(seed);

            double lowWeight  = -9d;
            double highWeight =  9d;

            int totalWeightCount = DeepNeuralNetwork.GetTotalWeightCount(inputNodeCount, hiddenNodeCountArray, outputNodeCount);

            double[] weightArray = new double[totalWeightCount];

            for(int i = 0; i < totalWeightCount; ++i)
            {
                weightArray[i] = (highWeight - lowWeight) * random.NextDouble() + lowWeight;
            }

            dnn.SetWeightArray(weightArray);

            double[][] resultArray = new double[itemCount][];

            for(int i = 0; i < itemCount; ++i)
            {
                resultArray[i] = new double[inputNodeCount + outputNodeCount];
            }

            double lowInputValue  = -4d;
            double highInputValue =  4d;

            for(int i = 0; i < itemCount; ++i)
            {
                double[] inputValueArray = new double[inputNodeCount];

                for(int j = 0; j < inputNodeCount; ++j)
                {
                    inputValueArray[j] = (highInputValue - lowInputValue) * random.NextDouble() + lowInputValue;
                }

                double[] outputValueArray          = dnn.Query(inputValueArray);
                double[] convertedOutputValueArray = ConvertOutputValueArray(outputValueArray);

                int k = 0;

                for(int j = 0; j < inputNodeCount; ++j)
                {
                    resultArray[i][k++] = inputValueArray[j];
                }

                for(int j = 0; j < outputNodeCount; ++j)
                {
                    resultArray[i][k++] = convertedOutputValueArray[j];
                }
            }

            return resultArray;
        }

        #endregion
        #region 훈련 데이터 배열 표시하기 - DisplayTrainingDataArray(sourceArray, rowCount, decimalPointCount, showIndex)

        /// <summary>
        /// 훈련 데이터 배열 표시하기
        /// </summary>
        /// <param name="sourceArray">소스 배열</param>
        /// <param name="rowCount">행 수</param>
        /// <param name="decimalPointCount">소수점 자리수</param>
        /// <param name="showIndex">인덱스 표시 여부</param>
        public static void DisplayTrainingDataArray(double[][] sourceArray, int rowCount, int decimalPointCount, bool showIndex)
        {
            int    sourceArrayLength             = sourceArray.Length;
            int    sourceArrayLengthStringLength = sourceArray.Length.ToString().Length;
            string displayFormat                 = string.Format("F{0}", decimalPointCount);

            for(int i = 0; i < rowCount; ++i)
            {
                if(showIndex == true)
                {
                    Console.Write("[{0}]  ", i.ToString().PadLeft(sourceArrayLengthStringLength));
                }

                for(int j = 0; j < sourceArray[i].Length; ++j)
                {
                    double value = sourceArray[i][j];

                    if(value >= 0d)
                    {
                        Console.Write(" ");
                    }

                    Console.Write(value.ToString(displayFormat) + "  ");
                }

                Console.WriteLine();
            }

            if(rowCount < sourceArrayLength)
            {
                Console.WriteLine("...");

                int lastRow = sourceArrayLength - 1;

                if(showIndex == true)
                {
                    Console.Write("[{0}]  ", lastRow.ToString().PadLeft(sourceArrayLengthStringLength));
                }

                for(int j = 0; j < sourceArray[lastRow].Length; ++j)
                {
                    double value = sourceArray[lastRow][j];

                    if(value >= 0d)
                    {
                        Console.Write(" ");
                    }

                    Console.Write(value.ToString(displayFormat) + "  ");
                }
            }

            Console.WriteLine();
        }

        #endregion
    }
}

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

using System;

namespace TestProject

{

/// <summary>

/// 프로그램

/// </summary>

class Program

{

//////////////////////////////////////////////////////////////////////////////////////////////////// Method

////////////////////////////////////////////////////////////////////////////////////////// Static

//////////////////////////////////////////////////////////////////////////////// Private

#region 프로그램 시작하기 - Main()

/// <summary>

/// 프로그램 시작하기

/// </summary>

private static void Main()

{

Console.WriteLine();

Console.WriteLine("심층 신경망(Deep Neural Network) 예제를 시작한다.");

Console.WriteLine();

int inputNodeCount = 4;

int[] hiddenNodeCountArray = new int[] { 10, 10, 10 };

int outputNodeCount = 3;

int dataItemCount = 2000;

Console.WriteLine("인공지능 훈련 데이터 항목 생성 시작 : {0}건", dataItemCount);

double[][] trainingDataArray = GetTrainingDataArray(dataItemCount, inputNodeCount, hiddenNodeCountArray, outputNodeCount, 5);

Console.WriteLine();

Console.WriteLine("인공지능 훈련 데이터 항목 생성 완료");

Console.WriteLine();

Console.WriteLine("인공지능 훈련 데이터 항목 : ");

Console.WriteLine();

DisplayTrainingDataArray(trainingDataArray, 3, 2, true);

Console.WriteLine();

Console.WriteLine("4-(10,10,10)-3 심층 신경망을 생성한다.");

Console.WriteLine();

DeepNeuralNetwork dnn = new DeepNeuralNetwork(inputNodeCount, hiddenNodeCountArray, outputNodeCount);

int maximumEpochCount = 2000;

double learningRate = 0.001;

double momentum = 0.01;

Console.WriteLine("최대 반복 차수 : {0}" , maximumEpochCount);

Console.WriteLine("학습률 : {0:F3}", learningRate );

Console.WriteLine("모멘텀 : {0:F3}", momentum);

Console.WriteLine();

Console.WriteLine("평균 제곱 오차를 갖는 역전파를 사용해 훈련을 시작한다.");

Console.WriteLine();

double[] bestWeightArray = dnn.Train(trainingDataArray, maximumEpochCount, learningRate, momentum, 10);

Console.WriteLine("훈련 완료.");

Console.WriteLine();

double trainingError = dnn.GetError(trainingDataArray, false);

double trainingAccuracy = dnn.GetAccuracy(trainingDataArray, false);

Console.WriteLine("최종 모델 MS 오차 : {0:F4}", trainingError );

Console.WriteLine("최종 모델 정확도 : {0:F4}", trainingAccuracy);

Console.ReadLine();

}

#endregion

#region 최대 값 인덱스 구하기 - GetMaximumValueIndex(sourceArray)

/// <summary>

/// 최대 값 인덱스 구하기

/// </summary>

/// <param name="sourceArray">소스 배열</param>

/// <returns>최대 값 인덱스</returns>

private static int GetMaximumValueIndex(double[] sourceArray)

{

int sourceArrayLength = sourceArray.Length;

int maximumValueIndex = 0;

double maximumValue = sourceArray[0];

for(int i = 0; i < sourceArrayLength; ++i)

{

if(sourceArray[i] > maximumValue)

{

maximumValue = sourceArray[i];

maximumValueIndex = i;

}

return maximumValueIndex;

}

#endregion

#region 결과 값 배열 변환하기 - ConvertOutputValueArray(sourceArray)

/// <summary>

/// 결과 값 배열 변환하기

/// </summary>

/// <param name="sourceArray">소스 배열</param>

/// <returns>변환 결과 값 배열</returns>

private static double[] ConvertOutputValueArray(double[] sourceArray)

{

double[] resultArray = new double[sourceArray.Length];

int maximumValueIndex = GetMaximumValueIndex(sourceArray);

resultArray[maximumValueIndex] = 1d;

return resultArray;

}

#endregion

#region 훈련 데이터 배열 구하기 - GetTrainingDataArray(itemCount, inputNodeCount, hiddenNodeCountArray, outputNodeCount, seed)

/// <summary>

/// 훈련 데이터 배열 구하기

/// </summary>

/// <param name="itemCount">항목 수</param>

/// <param name="inputNodeCount">입력 노드 수</param>

/// <param name="hiddenNodeCountArray">은닉 노드 수 배열</param>

/// <param name="outputNodeCount">출력 노드 수</param>

/// <param name="seed">시드</param>

/// <returns>훈련 데이터 배열</returns>

private static double[][] GetTrainingDataArray(int itemCount, int inputNodeCount, int[] hiddenNodeCountArray, int outputNodeCount, int seed)

{

DeepNeuralNetwork dnn = new DeepNeuralNetwork(inputNodeCount, hiddenNodeCountArray, outputNodeCount);

Random random = new Random(seed);

double lowWeight = -9d;

double highWeight = 9d;

int totalWeightCount = DeepNeuralNetwork.GetTotalWeightCount(inputNodeCount, hiddenNodeCountArray, outputNodeCount);

double[] weightArray = new double[totalWeightCount];

for(int i = 0; i < totalWeightCount; ++i)

{

weightArray[i] = (highWeight - lowWeight) * random.NextDouble() + lowWeight;

}

dnn.SetWeightArray(weightArray);

double[][] resultArray = new double[itemCount][];

for(int i = 0; i < itemCount; ++i)

{

resultArray[i] = new double[inputNodeCount + outputNodeCount];

}

double lowInputValue = -4d;

double highInputValue = 4d;

for(int i = 0; i < itemCount; ++i)

{

double[] inputValueArray = new double[inputNodeCount];

for(int j = 0; j < inputNodeCount; ++j)

{

inputValueArray[j] = (highInputValue - lowInputValue) * random.NextDouble() + lowInputValue;

}

double[] outputValueArray = dnn.Query(inputValueArray);

double[] convertedOutputValueArray = ConvertOutputValueArray(outputValueArray);

int k = 0;

for(int j = 0; j < inputNodeCount; ++j)

{

resultArray[i][k++] = inputValueArray[j];

}

for(int j = 0; j < outputNodeCount; ++j)

{

resultArray[i][k++] = convertedOutputValueArray[j];

}

return resultArray;

}

#endregion

#region 훈련 데이터 배열 표시하기 - DisplayTrainingDataArray(sourceArray, rowCount, decimalPointCount, showIndex)

/// <summary>

/// 훈련 데이터 배열 표시하기

/// </summary>

/// <param name="sourceArray">소스 배열</param>

/// <param name="rowCount">행 수</param>

/// <param name="decimalPointCount">소수점 자리수</param>

/// <param name="showIndex">인덱스 표시 여부</param>

public static void DisplayTrainingDataArray(double[][] sourceArray, int rowCount, int decimalPointCount, bool showIndex)

{

int sourceArrayLength = sourceArray.Length;

int sourceArrayLengthStringLength = sourceArray.Length.ToString().Length;

string displayFormat = string.Format("F{0}", decimalPointCount);

for(int i = 0; i < rowCount; ++i)

{

if(showIndex == true)

{

Console.Write("[{0}] ", i.ToString().PadLeft(sourceArrayLengthStringLength));

}

for(int j = 0; j < sourceArray[i].Length; ++j)

{

double value = sourceArray[i][j];

if(value >= 0d)

{

Console.Write(" ");

}

Console.Write(value.ToString(displayFormat) + " ");

}

Console.WriteLine();

}

if(rowCount < sourceArrayLength)

{

Console.WriteLine("...");

int lastRow = sourceArrayLength - 1;

if(showIndex == true)

{

Console.Write("[{0}] ", lastRow.ToString().PadLeft(sourceArrayLengthStringLength));

}

for(int j = 0; j < sourceArray[lastRow].Length; ++j)

{

double value = sourceArray[lastRow][j];

if(value >= 0d)

{

Console.Write(" ");

}

Console.Write(value.ToString(displayFormat) + " ");

}

Console.WriteLine();

}

#endregion

}

TestProject.zip

[C#/COMMON] 신경망 역전파 알고리즘 사용하기

■ 신경망 역전파 알고리즘을 사용하는 방법을 보여준다. ▶ NeuralNetwork.cs


using System;

namespace TestProject
{
    /// <summary>
    /// 신경망
    /// </summary>
    public class NeuralNetwork
    {
        //////////////////////////////////////////////////////////////////////////////////////////////////// Field
        ////////////////////////////////////////////////////////////////////////////////////////// Static
        //////////////////////////////////////////////////////////////////////////////// Private

        #region Field

        /// <summary>
        /// 난수 발생기
        /// </summary>
        private static Random _random;

        #endregion

        ////////////////////////////////////////////////////////////////////////////////////////// Instance
        //////////////////////////////////////////////////////////////////////////////// Private

        #region Field

        /// <summary>
        /// 입력 노드 카운트
        /// </summary>
        private int inputNodeCount;

        /// <summary>
        /// 은닉 노드 카운트
        /// </summary>
        private int hiddenNodeCount;

        /// <summary>
        /// 출력 노드 카운트
        /// </summary>
        private int outputNodeCount;


        /// <summary>
        /// 입력 노드 값 배열
        /// </summary>
        private double[] inputNodeValueArray;

        /// <summary>
        /// 입력-은닉 가중치 배열
        /// </summary>
        private double[][] inputHiddenWeightArray;

        /// <summary>
        /// 은닉 노드 바이어스 배열
        /// </summary>
        private double[] hiddenNodeBiasArray;

        /// <summary>
        /// 은닉 노드 값 배열
        /// </summary>
        private double[] hiddenNodeValueArray;

        /// <summary>
        /// 은닉-출력 가중치 배열
        /// </summary>
        private double[][] hiddenOutputWeightArray;

        /// <summary>
        /// 출력 노드 바이어스 배열
        /// </summary>
        private double[] outputNodeBiasArray;

        /// <summary>
        /// 출력 노드 값 배열
        /// </summary>
        private double[] outputNodeValueArray;


        /// <summary>
        /// 은닉 노드 그라디언트 배열
        /// </summary>
        private double[] hiddenNodeGradientArray;

        /// <summary>
        /// 출력 노드 그라디언트 배열
        /// </summary>
        private double[] outputNodeGradientArray;


        /// <summary>
        /// 입력-은닉 이전 가중치 델타 배열
        /// </summary>
        private double[][] inputHiddenPreviousWeightDeltaArray;

        /// <summary>
        /// 은닉 노드 이전 바이어스 델타 배열
        /// </summary>
        private double[] hiddenNodePreviousBiasDeltaArray;

        /// <summary>
        /// 은닉-출력 이전 가중치 델타 배열
        /// </summary>
        private double[][] hiddenOutputPreviousWeightDeltaArray;

        /// <summary>
        /// 출력 노드 이전 바이어스 델타 배열
        /// </summary>
        private double[] outputNodePreviousBiasDeltaArray;

        #endregion

        //////////////////////////////////////////////////////////////////////////////////////////////////// Constructor
        ////////////////////////////////////////////////////////////////////////////////////////// Public

        #region 생성자 - NeuralNetwork(inputNodeCount, hiddenNodeCount, outputNodeCount)

        /// <summary>
        /// 생성자
        /// </summary>
        /// <param name="inputNodeCount">입력 노드 카운트</param>
        /// <param name="hiddenNodeCount">은닉 노드 카운트</param>
        /// <param name="outputNodeCount">출력 노드 카운트</param>
        public NeuralNetwork(int inputNodeCount, int hiddenNodeCount, int outputNodeCount)
        {
            _random = new Random(0);

            this.inputNodeCount  = inputNodeCount;
            this.hiddenNodeCount = hiddenNodeCount;
            this.outputNodeCount = outputNodeCount;

            this.inputNodeValueArray     = new double[inputNodeCount];
            this.inputHiddenWeightArray  = GetMatrixArray(inputNodeCount, hiddenNodeCount);
            this.hiddenNodeBiasArray     = new double[hiddenNodeCount];
            this.hiddenNodeValueArray    = new double[hiddenNodeCount];
            this.hiddenOutputWeightArray = GetMatrixArray(hiddenNodeCount, outputNodeCount);
            this.outputNodeBiasArray     = new double[outputNodeCount];
            this.outputNodeValueArray    = new double[outputNodeCount];

            this.InitializeWeightArray();

            this.hiddenNodeGradientArray = new double[hiddenNodeCount];
            this.outputNodeGradientArray = new double[outputNodeCount];

            this.inputHiddenPreviousWeightDeltaArray  = GetMatrixArray(inputNodeCount, hiddenNodeCount);
            this.hiddenNodePreviousBiasDeltaArray     = new double[hiddenNodeCount];
            this.hiddenOutputPreviousWeightDeltaArray = GetMatrixArray(hiddenNodeCount, outputNodeCount);
            this.outputNodePreviousBiasDeltaArray     = new double[outputNodeCount];
        }

        #endregion

        //////////////////////////////////////////////////////////////////////////////////////////////////// Method
        ////////////////////////////////////////////////////////////////////////////////////////// Public

        #region 가중치 배열 구하기 - GetWeightArray()

        /// <summary>
        /// 가중치 배열 구하기
        /// </summary>
        /// <returns>가중치 배열</returns>
        public double[] GetWeightArray()
        {
            int totalWeightCount = (this.inputNodeCount  * this.hiddenNodeCount) +
                                   (this.hiddenNodeCount * this.outputNodeCount) +
                                    this.hiddenNodeCount + this.outputNodeCount;

            double[] targetWeightArray = new double[totalWeightCount];

            int k = 0;

            for(int i = 0; i < this.inputNodeCount; i++)
            {
                for(int j = 0; j < this.hiddenNodeCount; j++)
                {
                    targetWeightArray[k++] = this.inputHiddenWeightArray[i][j];
                }
            }

            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                targetWeightArray[k++] = this.hiddenNodeBiasArray[i];
            }

            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                for(int j = 0; j < this.outputNodeCount; j++)
                {
                    targetWeightArray[k++] = this.hiddenOutputWeightArray[i][j];
                }
            }

            for(int i = 0; i < this.outputNodeCount; i++)
            {
                targetWeightArray[k++] = this.outputNodeBiasArray[i];
            }

            return targetWeightArray;
        }

        #endregion
        #region 가중치 배열 설정하기 - SetWeightArray(sourceWeightArray)

        /// <summary>
        /// 가중치 배열 설정하기
        /// </summary>
        /// <param name="sourceWeightArray">소스 가중치 배열</param>
        public void SetWeightArray(double[] sourceWeightArray)
        {
            int totalWeightCount = (this.inputNodeCount  * this.hiddenNodeCount) +
                                   (this.hiddenNodeCount * this.outputNodeCount) +
                                    this.hiddenNodeCount + this.outputNodeCount;

            if(sourceWeightArray.Length != totalWeightCount)
            {
                throw new Exception("잘못된 소스 가중치 배열 길이 입니다.");
            }

            int k = 0;

            for(int i = 0; i < this.inputNodeCount; i++)
            {
                for(int j = 0; j < this.hiddenNodeCount; j++)
                {
                    this.inputHiddenWeightArray[i][j] = sourceWeightArray[k++];
                }
            }

            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                this.hiddenNodeBiasArray[i] = sourceWeightArray[k++];
            }

            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                for(int j = 0; j < this.outputNodeCount; j++)
                {
                    this.hiddenOutputWeightArray[i][j] = sourceWeightArray[k++];
                }
            }

            for(int i = 0; i < this.outputNodeCount; i++)
            {
                this.outputNodeBiasArray[i] = sourceWeightArray[k++];
            }
        }

        #endregion
        #region 학습하기 - Train(trainingValueArray, maximumEpochCount, learningRate, momentum)

        /// <summary>
        /// 학습하기
        /// </summary>
        /// <param name="trainingValueArray">학습 값 배열</param>
        /// <param name="maximumEpochCount">최대 시대 카운트</param>
        /// <param name="learningRate">학습률</param>
        /// <param name="momentum">모멘텀</param>
        public void Train(double[][] trainingValueArray, int maximumEpochCount, double learningRate, double momentum)
        {
            int      epochCount          = 0;
            double[] xValueArray         = new double[this.inputNodeCount ];
            double[] targetValueArray    = new double[this.outputNodeCount];
            int      traingingValueCount = trainingValueArray.Length;
            int[]    sequenceArray       = new int[traingingValueCount];
            int      sequenceCount       = sequenceArray.Length;

            for(int i = 0; i < sequenceCount; ++i)
            {
                sequenceArray[i] = i;
            }

            while(epochCount < maximumEpochCount)
            {
                double mse = GetMeanSquaredError(trainingValueArray);

                if(mse < 0.04d)
                {
                    break;
                }

                Shuffle(sequenceArray);

                for(int i = 0; i < traingingValueCount; i++)
                {
                    int sequenceIndex = sequenceArray[i];

                    Array.Copy(trainingValueArray[sequenceIndex], xValueArray, this.inputNodeCount);

                    Array.Copy(trainingValueArray[sequenceIndex], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                    ComputeOutputNodeValueArray(xValueArray);

                    UpdateWeightArray(targetValueArray, learningRate, momentum);
                }

                epochCount++;
            }
        }

        #endregion
        #region 출력 노드 배열 계산하기 - ComputeOutputNodeValueArray(sourceInputNodeValueArray)

        /// <summary>
        /// 출력 노드 배열 계산하기
        /// </summary>
        /// <param name="sourceInputNodeValueArray">소스 입력 노드 값 배열</param>
        /// <returns>출력 노드 값 배열</returns>
        public double[] ComputeOutputNodeValueArray(double[] sourceInputNodeValueArray)
        {
            if(sourceInputNodeValueArray.Length != this.inputNodeCount)
            {
                throw new Exception("잘못된 소스 입력 노드 값 배열 길이 입니다.");
            }

            double[] hiddenNodeSummaryArray = new double[this.hiddenNodeCount];
            double[] outputNodeSummaryArray = new double[this.outputNodeCount];

            int sourceInputNodeValueCount = sourceInputNodeValueArray.Length;

            for(int i = 0; i < sourceInputNodeValueCount; i++)
            {
                this.inputNodeValueArray[i] = sourceInputNodeValueArray[i];
            }

            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                for(int j = 0; j < this.inputNodeCount; j++)
                {
                    hiddenNodeSummaryArray[i] += this.inputNodeValueArray[j] * this.inputHiddenWeightArray[j][i];
                }
            }

            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                hiddenNodeSummaryArray[i] += this.hiddenNodeBiasArray[i];
            }

            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                this.hiddenNodeValueArray[i] = GetHyperTangentValue(hiddenNodeSummaryArray[i]);
            }

            for(int i = 0; i < this.outputNodeCount; i++)
            {
                for(int j = 0; j < this.hiddenNodeCount; j++)
                {
                    outputNodeSummaryArray[i] += this.hiddenNodeValueArray[j] * this.hiddenOutputWeightArray[j][i];
                }
            }

            for(int i = 0; i < this.outputNodeCount; i++)
            {
                outputNodeSummaryArray[i] += this.outputNodeBiasArray[i];
            }

            double[] softMaximumArray = GetSoftMaximumArray(outputNodeSummaryArray);

            Array.Copy(softMaximumArray, this.outputNodeValueArray, softMaximumArray.Length);

            double[] targetOutputNodeValueArray = new double[this.outputNodeCount];

            Array.Copy(this.outputNodeValueArray, targetOutputNodeValueArray, targetOutputNodeValueArray.Length);

            return targetOutputNodeValueArray;
        }

        #endregion
        #region 정확도 구하기 - GetAccuracy(trainingValueArray)

        /// <summary>
        /// 정확도 구하기
        /// </summary>
        /// <param name="trainingValueArray">훈련 값 배열</param>
        /// <returns>정확도</returns>
        public double GetAccuracy(double[][] trainingValueArray)
        {
            int      correctCount       = 0;
            int      wrongCount         = 0;
            double[] xValueArray        = new double[this.inputNodeCount ];
            double[] targetValueArray   = new double[this.outputNodeCount];
            double[] yValueArray;
            int      trainingValueCount = trainingValueArray.Length;

            for(int i = 0; i < trainingValueCount; i++)
            {
                Array.Copy(trainingValueArray[i], xValueArray, this.inputNodeCount);

                Array.Copy(trainingValueArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                yValueArray = ComputeOutputNodeValueArray(xValueArray);

                int maximumIndex = GetMaximumIndex(yValueArray);

                if(targetValueArray[maximumIndex] == 1d)
                {
                    correctCount++;
                }
                else
                {
                    wrongCount++;
                }
            }

            return (correctCount * 1d) / (correctCount + wrongCount);
        }

        #endregion

        ////////////////////////////////////////////////////////////////////////////////////////// Private

        #region 매트릭스 배열 구하기 - GetMatrixArray(rowCount, columnCount)

        /// <summary>
        /// 매트릭스 배열 구하기
        /// </summary>
        /// <param name="rowCount">행 카운트</param>
        /// <param name="columnCount">컬럼 카운트</param>
        /// <returns>매트릭스 배열</returns>
        private static double[][] GetMatrixArray(int rowCount, int columnCount)
        {
            double[][] matrixArray = new double[rowCount][];

            int count = matrixArray.Length;

            for(int i = 0; i < count; i++)
            {
                matrixArray[i] = new double[columnCount];
            }

            return matrixArray;
        }

        #endregion
        #region 가중치 배열 초기화 하기 - InitializeWeightArray()

        /// <summary>
        /// 가중치 배열 초기화 하기
        /// </summary>
        private void InitializeWeightArray()
        {
            int totalWeightCount = (this.inputNodeCount  * this.hiddenNodeCount) +
                                   (this.hiddenNodeCount * this.outputNodeCount) +
                                    this.hiddenNodeCount + this.outputNodeCount;

            double[] initialWeightArray = new double[totalWeightCount];
            double   low                = -0.01d;
            double   high               =  0.01d;
            int      initialWeightCount = initialWeightArray.Length;

            for(int i = 0; i < initialWeightCount; i++)
            {
                initialWeightArray[i] = (high - low) * _random.NextDouble() + low;
            }

            SetWeightArray(initialWeightArray);
        }

        #endregion
        #region 하이퍼 탄젠트 값 구하기 - GetHyperTangentValue(x)

        /// <summary>
        /// 하이퍼 탄젠트 값 구하기
        /// </summary>
        /// <param name="x">X</param>
        /// <returns>하이퍼 탄젠트 값</returns>
        private static double GetHyperTangentValue(double x)
        {
            if(x < -20d)
            {
                return -1d;
            }
            else if(x > 20d)
            {
                return 1d;
            }
            else
            {
                return Math.Tanh(x);
            }
        }

        #endregion
        #region SOFT-MAX 배열 구하기 - GetSoftMaximumArray(sourceOutputNodeSummaryArray)

        /// <summary>
        /// SOFT-MAX 배열 구하기
        /// </summary>
        /// <param name="sourceOutputNodeSummaryArray">소스 출력 노드 합산 배열</param>
        /// <returns>SOFT-MAX 배열</returns>
        private static double[] GetSoftMaximumArray(double[] sourceOutputNodeSummaryArray)
        {
            int sourceOutputNodeSummaryCount = sourceOutputNodeSummaryArray.Length;

            double maximum = sourceOutputNodeSummaryArray[0];

            for(int i = 0; i < sourceOutputNodeSummaryCount; i++)
            {
                if(sourceOutputNodeSummaryArray[i] > maximum)
                {
                    maximum = sourceOutputNodeSummaryArray[i];
                }
            }

            double scale = 0d;

            for(int i = 0; i < sourceOutputNodeSummaryCount; ++i)
            {
                scale += Math.Exp(sourceOutputNodeSummaryArray[i] - maximum);
            }

            double[] targetArray = new double[sourceOutputNodeSummaryCount];

            for(int i = 0; i < sourceOutputNodeSummaryCount; ++i)
            {
                targetArray[i] = Math.Exp(sourceOutputNodeSummaryArray[i] - maximum) / scale;
            }

            return targetArray;
        }

        #endregion
        #region 가중치 배열 갱신하기 - UpdateWeightArray(targetValueArray, learningRate, momentum)

        /// <summary>
        /// 가중치 배열 갱신하기
        /// </summary>
        /// <param name="targetValueArray">타겟 값 배열</param>
        /// <param name="learningRate">학습률</param>
        /// <param name="momentum">모멘텀</param>
        private void UpdateWeightArray(double[] targetValueArray, double learningRate, double momentum)
        {
            if(targetValueArray.Length != this.outputNodeCount)
            {
                throw new Exception("타겟 값 배열 길이가 출력 노드 값 배열과 같지 않습니다.");
            }


            for(int i = 0; i < this.outputNodeCount; i++)
            {
                double derivative = (1 - this.outputNodeValueArray[i]) * this.outputNodeValueArray[i];

                this.outputNodeGradientArray[i] = derivative * (targetValueArray[i] - this.outputNodeValueArray[i]);
            }


            for(int i = 0; i < hiddenNodeCount; i++)
            {
                double derivative = (1 - this.hiddenNodeValueArray[i]) * (1 + this.hiddenNodeValueArray[i]);
                double summary    = 0d;

                for(int j = 0; j < this.outputNodeCount; j++)
                {
                    double x = this.outputNodeGradientArray[j] * this.hiddenOutputWeightArray[i][j];

                    summary += x;
                }

                this.hiddenNodeGradientArray[i] = derivative * summary;
            }


            for(int i = 0; i < this.inputNodeCount; i++)
            {
                for(int j = 0; j < this.hiddenNodeCount; j++)
                {
                    double delta = learningRate * this.hiddenNodeGradientArray[j] * this.inputNodeValueArray[i];

                    this.inputHiddenWeightArray[i][j] += delta;

                    this.inputHiddenWeightArray[i][j] += momentum * this.inputHiddenPreviousWeightDeltaArray[i][j];

                    this.inputHiddenPreviousWeightDeltaArray[i][j] = delta;
                }
            }


            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                double delta = learningRate * this.hiddenNodeGradientArray[i];

                this.hiddenNodeBiasArray[i] += delta;
                this.hiddenNodeBiasArray[i] += momentum * this.hiddenNodePreviousBiasDeltaArray[i];

                this.hiddenNodePreviousBiasDeltaArray[i] = delta;
            }


            for(int i = 0; i < this.hiddenNodeCount; i++)
            {
                for(int j = 0; j < this.outputNodeCount; j++)
                {
                    double delta = learningRate * this.outputNodeGradientArray[j] * this.hiddenNodeValueArray[i];

                    this.hiddenOutputWeightArray[i][j] += delta;
                    this.hiddenOutputWeightArray[i][j] += momentum * this.hiddenOutputPreviousWeightDeltaArray[i][j];

                    this.hiddenOutputPreviousWeightDeltaArray[i][j] = delta;
                }
            }


            for(int i = 0; i < this.outputNodeCount; i++)
            {
                double delta = learningRate * this.outputNodeGradientArray[i] * 1d;

                this.outputNodeBiasArray[i] += delta;
                this.outputNodeBiasArray[i] += momentum * this.outputNodePreviousBiasDeltaArray[i];

                this.outputNodePreviousBiasDeltaArray[i] = delta;
            }
        }

        #endregion
        #region 섞기 - Shuffle(sequenceArray)

        /// <summary>
        /// 섞기
        /// </summary>
        /// <param name="sequenceArray">시퀀스 배열</param>
        private static void Shuffle(int[] sequenceArray)
        {
            for(int i = 0; i < sequenceArray.Length; i++)
            {
                int randomIndex = _random.Next(i, sequenceArray.Length);
                int sequence    = sequenceArray[randomIndex];

                sequenceArray[randomIndex] = sequenceArray[i];

                sequenceArray[i] = sequence;
            }
        }

        #endregion
        #region 평균 제곱 에러 구하기 - GetMeanSquaredError(trainingValueArray)

        /// <summary>
        /// 평균 제곱 에러 구하기
        /// </summary>
        /// <param name="trainingValueArray">훈련 값 배열</param>
        /// <returns>평균 제곱 에러</returns>
        private double GetMeanSquaredError(double[][] trainingValueArray)
        {
            double   summarySquaredError = 0d;
            double[] xValueArray         = new double[this.inputNodeCount ];
            double[] targetValueArray    = new double[this.outputNodeCount];
            int      trainingValueCount  = trainingValueArray.Length;

            for(int i = 0; i < trainingValueCount; i++)
            {
                Array.Copy(trainingValueArray[i], xValueArray, this.inputNodeCount);

                Array.Copy(trainingValueArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

                double[] yValueArray = ComputeOutputNodeValueArray(xValueArray);

                for(int j = 0; j < this.outputNodeCount; j++)
                {
                    double error = targetValueArray[j] - yValueArray[j];

                    summarySquaredError += error * error;
                }
            }

            return summarySquaredError / trainingValueCount;
        }

        #endregion
        #region 최대 인덱스 구하기 - GetMaximumIndex(vectorArray)

        /// <summary>
        /// 최대 인덱스 구하기
        /// </summary>
        /// <param name="vectorArray">벡터 배열</param>
        /// <returns>최대 인덱스</returns>
        private static int GetMaximumIndex(double[] vectorArray)
        {
            int    maximumIndex = 0;
            double maximum      = vectorArray[0];
            int    vectorCount  = vectorArray.Length;

            for(int i = 0; i < vectorCount; i++)
            {
                if(vectorArray[i] > maximum)
                {
                    maximum      = vectorArray[i];
                    maximumIndex = i;
                }
            }

            return maximumIndex;
        }

        #endregion
    }
}

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

using System;

namespace TestProject

{

/// <summary>

/// 신경망

/// </summary>

public class NeuralNetwork

{

//////////////////////////////////////////////////////////////////////////////////////////////////// Field

////////////////////////////////////////////////////////////////////////////////////////// Static

//////////////////////////////////////////////////////////////////////////////// Private

#region Field

/// <summary>

/// 난수 발생기

/// </summary>

private static Random _random;

#endregion

////////////////////////////////////////////////////////////////////////////////////////// Instance

//////////////////////////////////////////////////////////////////////////////// Private

#region Field

/// <summary>

/// 입력 노드 카운트

/// </summary>

private int inputNodeCount;

/// <summary>

/// 은닉 노드 카운트

/// </summary>

private int hiddenNodeCount;

/// <summary>

/// 출력 노드 카운트

/// </summary>

private int outputNodeCount;

/// <summary>

/// 입력 노드 값 배열

/// </summary>

private double[] inputNodeValueArray;

/// <summary>

/// 입력-은닉 가중치 배열

/// </summary>

private double[][] inputHiddenWeightArray;

/// <summary>

/// 은닉 노드 바이어스 배열

/// </summary>

private double[] hiddenNodeBiasArray;

/// <summary>

/// 은닉 노드 값 배열

/// </summary>

private double[] hiddenNodeValueArray;

/// <summary>

/// 은닉-출력 가중치 배열

/// </summary>

private double[][] hiddenOutputWeightArray;

/// <summary>

/// 출력 노드 바이어스 배열

/// </summary>

private double[] outputNodeBiasArray;

/// <summary>

/// 출력 노드 값 배열

/// </summary>

private double[] outputNodeValueArray;

/// <summary>

/// 은닉 노드 그라디언트 배열

/// </summary>

private double[] hiddenNodeGradientArray;

/// <summary>

/// 출력 노드 그라디언트 배열

/// </summary>

private double[] outputNodeGradientArray;

/// <summary>

/// 입력-은닉 이전 가중치 델타 배열

/// </summary>

private double[][] inputHiddenPreviousWeightDeltaArray;

/// <summary>

/// 은닉 노드 이전 바이어스 델타 배열

/// </summary>

private double[] hiddenNodePreviousBiasDeltaArray;

/// <summary>

/// 은닉-출력 이전 가중치 델타 배열

/// </summary>

private double[][] hiddenOutputPreviousWeightDeltaArray;

/// <summary>

/// 출력 노드 이전 바이어스 델타 배열

/// </summary>

private double[] outputNodePreviousBiasDeltaArray;

#endregion

//////////////////////////////////////////////////////////////////////////////////////////////////// Constructor

////////////////////////////////////////////////////////////////////////////////////////// Public

#region 생성자 - NeuralNetwork(inputNodeCount, hiddenNodeCount, outputNodeCount)

/// <summary>

/// 생성자

/// </summary>

/// <param name="inputNodeCount">입력 노드 카운트</param>

/// <param name="hiddenNodeCount">은닉 노드 카운트</param>

/// <param name="outputNodeCount">출력 노드 카운트</param>

public NeuralNetwork(int inputNodeCount, int hiddenNodeCount, int outputNodeCount)

{

_random = new Random(0);

this.inputNodeCount = inputNodeCount;

this.hiddenNodeCount = hiddenNodeCount;

this.outputNodeCount = outputNodeCount;

this.inputNodeValueArray = new double[inputNodeCount];

this.inputHiddenWeightArray = GetMatrixArray(inputNodeCount, hiddenNodeCount);

this.hiddenNodeBiasArray = new double[hiddenNodeCount];

this.hiddenNodeValueArray = new double[hiddenNodeCount];

this.hiddenOutputWeightArray = GetMatrixArray(hiddenNodeCount, outputNodeCount);

this.outputNodeBiasArray = new double[outputNodeCount];

this.outputNodeValueArray = new double[outputNodeCount];

this.InitializeWeightArray();

this.hiddenNodeGradientArray = new double[hiddenNodeCount];

this.outputNodeGradientArray = new double[outputNodeCount];

this.inputHiddenPreviousWeightDeltaArray = GetMatrixArray(inputNodeCount, hiddenNodeCount);

this.hiddenNodePreviousBiasDeltaArray = new double[hiddenNodeCount];

this.hiddenOutputPreviousWeightDeltaArray = GetMatrixArray(hiddenNodeCount, outputNodeCount);

this.outputNodePreviousBiasDeltaArray = new double[outputNodeCount];

}

#endregion

//////////////////////////////////////////////////////////////////////////////////////////////////// Method

////////////////////////////////////////////////////////////////////////////////////////// Public

#region 가중치 배열 구하기 - GetWeightArray()

/// <summary>

/// 가중치 배열 구하기

/// </summary>

/// <returns>가중치 배열</returns>

public double[] GetWeightArray()

{

int totalWeightCount = (this.inputNodeCount * this.hiddenNodeCount) +

(this.hiddenNodeCount * this.outputNodeCount) +

this.hiddenNodeCount + this.outputNodeCount;

double[] targetWeightArray = new double[totalWeightCount];

int k = 0;

for(int i = 0; i < this.inputNodeCount; i++)

{

for(int j = 0; j < this.hiddenNodeCount; j++)

{

targetWeightArray[k++] = this.inputHiddenWeightArray[i][j];

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

targetWeightArray[k++] = this.hiddenNodeBiasArray[i];

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

for(int j = 0; j < this.outputNodeCount; j++)

{

targetWeightArray[k++] = this.hiddenOutputWeightArray[i][j];

}

for(int i = 0; i < this.outputNodeCount; i++)

{

targetWeightArray[k++] = this.outputNodeBiasArray[i];

}

return targetWeightArray;

}

#endregion

#region 가중치 배열 설정하기 - SetWeightArray(sourceWeightArray)

/// <summary>

/// 가중치 배열 설정하기

/// </summary>

/// <param name="sourceWeightArray">소스 가중치 배열</param>

public void SetWeightArray(double[] sourceWeightArray)

{

int totalWeightCount = (this.inputNodeCount * this.hiddenNodeCount) +

(this.hiddenNodeCount * this.outputNodeCount) +

this.hiddenNodeCount + this.outputNodeCount;

if(sourceWeightArray.Length != totalWeightCount)

{

throw new Exception("잘못된 소스 가중치 배열 길이 입니다.");

}

int k = 0;

for(int i = 0; i < this.inputNodeCount; i++)

{

for(int j = 0; j < this.hiddenNodeCount; j++)

{

this.inputHiddenWeightArray[i][j] = sourceWeightArray[k++];

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

this.hiddenNodeBiasArray[i] = sourceWeightArray[k++];

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

for(int j = 0; j < this.outputNodeCount; j++)

{

this.hiddenOutputWeightArray[i][j] = sourceWeightArray[k++];

}

for(int i = 0; i < this.outputNodeCount; i++)

{

this.outputNodeBiasArray[i] = sourceWeightArray[k++];

}

#endregion

#region 학습하기 - Train(trainingValueArray, maximumEpochCount, learningRate, momentum)

/// <summary>

/// 학습하기

/// </summary>

/// <param name="trainingValueArray">학습 값 배열</param>

/// <param name="maximumEpochCount">최대 시대 카운트</param>

/// <param name="learningRate">학습률</param>

/// <param name="momentum">모멘텀</param>

public void Train(double[][] trainingValueArray, int maximumEpochCount, double learningRate, double momentum)

{

int epochCount = 0;

double[] xValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

int traingingValueCount = trainingValueArray.Length;

int[] sequenceArray = new int[traingingValueCount];

int sequenceCount = sequenceArray.Length;

for(int i = 0; i < sequenceCount; ++i)

{

sequenceArray[i] = i;

}

while(epochCount < maximumEpochCount)

{

double mse = GetMeanSquaredError(trainingValueArray);

if(mse < 0.04d)

{

break;

}

Shuffle(sequenceArray);

for(int i = 0; i < traingingValueCount; i++)

{

int sequenceIndex = sequenceArray[i];

Array.Copy(trainingValueArray[sequenceIndex], xValueArray, this.inputNodeCount);

Array.Copy(trainingValueArray[sequenceIndex], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

ComputeOutputNodeValueArray(xValueArray);

UpdateWeightArray(targetValueArray, learningRate, momentum);

}

epochCount++;

}

#endregion

#region 출력 노드 배열 계산하기 - ComputeOutputNodeValueArray(sourceInputNodeValueArray)

/// <summary>

/// 출력 노드 배열 계산하기

/// </summary>

/// <param name="sourceInputNodeValueArray">소스 입력 노드 값 배열</param>

/// <returns>출력 노드 값 배열</returns>

public double[] ComputeOutputNodeValueArray(double[] sourceInputNodeValueArray)

{

if(sourceInputNodeValueArray.Length != this.inputNodeCount)

{

throw new Exception("잘못된 소스 입력 노드 값 배열 길이 입니다.");

}

double[] hiddenNodeSummaryArray = new double[this.hiddenNodeCount];

double[] outputNodeSummaryArray = new double[this.outputNodeCount];

int sourceInputNodeValueCount = sourceInputNodeValueArray.Length;

for(int i = 0; i < sourceInputNodeValueCount; i++)

{

this.inputNodeValueArray[i] = sourceInputNodeValueArray[i];

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

for(int j = 0; j < this.inputNodeCount; j++)

{

hiddenNodeSummaryArray[i] += this.inputNodeValueArray[j] * this.inputHiddenWeightArray[j][i];

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

hiddenNodeSummaryArray[i] += this.hiddenNodeBiasArray[i];

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

this.hiddenNodeValueArray[i] = GetHyperTangentValue(hiddenNodeSummaryArray[i]);

}

for(int i = 0; i < this.outputNodeCount; i++)

{

for(int j = 0; j < this.hiddenNodeCount; j++)

{

outputNodeSummaryArray[i] += this.hiddenNodeValueArray[j] * this.hiddenOutputWeightArray[j][i];

}

for(int i = 0; i < this.outputNodeCount; i++)

{

outputNodeSummaryArray[i] += this.outputNodeBiasArray[i];

}

double[] softMaximumArray = GetSoftMaximumArray(outputNodeSummaryArray);

Array.Copy(softMaximumArray, this.outputNodeValueArray, softMaximumArray.Length);

double[] targetOutputNodeValueArray = new double[this.outputNodeCount];

Array.Copy(this.outputNodeValueArray, targetOutputNodeValueArray, targetOutputNodeValueArray.Length);

return targetOutputNodeValueArray;

}

#endregion

#region 정확도 구하기 - GetAccuracy(trainingValueArray)

/// <summary>

/// 정확도 구하기

/// </summary>

/// <param name="trainingValueArray">훈련 값 배열</param>

/// <returns>정확도</returns>

public double GetAccuracy(double[][] trainingValueArray)

{

int correctCount = 0;

int wrongCount = 0;

double[] xValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

double[] yValueArray;

int trainingValueCount = trainingValueArray.Length;

for(int i = 0; i < trainingValueCount; i++)

{

Array.Copy(trainingValueArray[i], xValueArray, this.inputNodeCount);

Array.Copy(trainingValueArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

yValueArray = ComputeOutputNodeValueArray(xValueArray);

int maximumIndex = GetMaximumIndex(yValueArray);

if(targetValueArray[maximumIndex] == 1d)

{

correctCount++;

}

else

{

wrongCount++;

}

return (correctCount * 1d) / (correctCount + wrongCount);

}

#endregion

////////////////////////////////////////////////////////////////////////////////////////// Private

#region 매트릭스 배열 구하기 - GetMatrixArray(rowCount, columnCount)

/// <summary>

/// 매트릭스 배열 구하기

/// </summary>

/// <param name="rowCount">행 카운트</param>

/// <param name="columnCount">컬럼 카운트</param>

/// <returns>매트릭스 배열</returns>

private static double[][] GetMatrixArray(int rowCount, int columnCount)

{

double[][] matrixArray = new double[rowCount][];

int count = matrixArray.Length;

for(int i = 0; i < count; i++)

{

matrixArray[i] = new double[columnCount];

}

return matrixArray;

}

#endregion

#region 가중치 배열 초기화 하기 - InitializeWeightArray()

/// <summary>

/// 가중치 배열 초기화 하기

/// </summary>

private void InitializeWeightArray()

{

int totalWeightCount = (this.inputNodeCount * this.hiddenNodeCount) +

(this.hiddenNodeCount * this.outputNodeCount) +

this.hiddenNodeCount + this.outputNodeCount;

double[] initialWeightArray = new double[totalWeightCount];

double low = -0.01d;

double high = 0.01d;

int initialWeightCount = initialWeightArray.Length;

for(int i = 0; i < initialWeightCount; i++)

{

initialWeightArray[i] = (high - low) * _random.NextDouble() + low;

}

SetWeightArray(initialWeightArray);

}

#endregion

#region 하이퍼 탄젠트 값 구하기 - GetHyperTangentValue(x)

/// <summary>

/// 하이퍼 탄젠트 값 구하기

/// </summary>

/// <param name="x">X</param>

/// <returns>하이퍼 탄젠트 값</returns>

private static double GetHyperTangentValue(double x)

{

if(x < -20d)

{

return -1d;

}

else if(x > 20d)

{

return 1d;

}

else

{

return Math.Tanh(x);

}

#endregion

#region SOFT-MAX 배열 구하기 - GetSoftMaximumArray(sourceOutputNodeSummaryArray)

/// <summary>

/// SOFT-MAX 배열 구하기

/// </summary>

/// <param name="sourceOutputNodeSummaryArray">소스 출력 노드 합산 배열</param>

/// <returns>SOFT-MAX 배열</returns>

private static double[] GetSoftMaximumArray(double[] sourceOutputNodeSummaryArray)

{

int sourceOutputNodeSummaryCount = sourceOutputNodeSummaryArray.Length;

double maximum = sourceOutputNodeSummaryArray[0];

for(int i = 0; i < sourceOutputNodeSummaryCount; i++)

{

if(sourceOutputNodeSummaryArray[i] > maximum)

{

maximum = sourceOutputNodeSummaryArray[i];

}

double scale = 0d;

for(int i = 0; i < sourceOutputNodeSummaryCount; ++i)

{

scale += Math.Exp(sourceOutputNodeSummaryArray[i] - maximum);

}

double[] targetArray = new double[sourceOutputNodeSummaryCount];

for(int i = 0; i < sourceOutputNodeSummaryCount; ++i)

{

targetArray[i] = Math.Exp(sourceOutputNodeSummaryArray[i] - maximum) / scale;

}

return targetArray;

}

#endregion

#region 가중치 배열 갱신하기 - UpdateWeightArray(targetValueArray, learningRate, momentum)

/// <summary>

/// 가중치 배열 갱신하기

/// </summary>

/// <param name="targetValueArray">타겟 값 배열</param>

/// <param name="learningRate">학습률</param>

/// <param name="momentum">모멘텀</param>

private void UpdateWeightArray(double[] targetValueArray, double learningRate, double momentum)

{

if(targetValueArray.Length != this.outputNodeCount)

{

throw new Exception("타겟 값 배열 길이가 출력 노드 값 배열과 같지 않습니다.");

}

for(int i = 0; i < this.outputNodeCount; i++)

{

double derivative = (1 - this.outputNodeValueArray[i]) * this.outputNodeValueArray[i];

this.outputNodeGradientArray[i] = derivative * (targetValueArray[i] - this.outputNodeValueArray[i]);

}

for(int i = 0; i < hiddenNodeCount; i++)

{

double derivative = (1 - this.hiddenNodeValueArray[i]) * (1 + this.hiddenNodeValueArray[i]);

double summary = 0d;

for(int j = 0; j < this.outputNodeCount; j++)

{

double x = this.outputNodeGradientArray[j] * this.hiddenOutputWeightArray[i][j];

summary += x;

}

this.hiddenNodeGradientArray[i] = derivative * summary;

}

for(int i = 0; i < this.inputNodeCount; i++)

{

for(int j = 0; j < this.hiddenNodeCount; j++)

{

double delta = learningRate * this.hiddenNodeGradientArray[j] * this.inputNodeValueArray[i];

this.inputHiddenWeightArray[i][j] += delta;

this.inputHiddenWeightArray[i][j] += momentum * this.inputHiddenPreviousWeightDeltaArray[i][j];

this.inputHiddenPreviousWeightDeltaArray[i][j] = delta;

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

double delta = learningRate * this.hiddenNodeGradientArray[i];

this.hiddenNodeBiasArray[i] += delta;

this.hiddenNodeBiasArray[i] += momentum * this.hiddenNodePreviousBiasDeltaArray[i];

this.hiddenNodePreviousBiasDeltaArray[i] = delta;

}

for(int i = 0; i < this.hiddenNodeCount; i++)

{

for(int j = 0; j < this.outputNodeCount; j++)

{

double delta = learningRate * this.outputNodeGradientArray[j] * this.hiddenNodeValueArray[i];

this.hiddenOutputWeightArray[i][j] += delta;

this.hiddenOutputWeightArray[i][j] += momentum * this.hiddenOutputPreviousWeightDeltaArray[i][j];

this.hiddenOutputPreviousWeightDeltaArray[i][j] = delta;

}

for(int i = 0; i < this.outputNodeCount; i++)

{

double delta = learningRate * this.outputNodeGradientArray[i] * 1d;

this.outputNodeBiasArray[i] += delta;

this.outputNodeBiasArray[i] += momentum * this.outputNodePreviousBiasDeltaArray[i];

this.outputNodePreviousBiasDeltaArray[i] = delta;

}

#endregion

#region 섞기 - Shuffle(sequenceArray)

/// <summary>

/// 섞기

/// </summary>

/// <param name="sequenceArray">시퀀스 배열</param>

private static void Shuffle(int[] sequenceArray)

{

for(int i = 0; i < sequenceArray.Length; i++)

{

int randomIndex = _random.Next(i, sequenceArray.Length);

int sequence = sequenceArray[randomIndex];

sequenceArray[randomIndex] = sequenceArray[i];

sequenceArray[i] = sequence;

}

#endregion

#region 평균 제곱 에러 구하기 - GetMeanSquaredError(trainingValueArray)

/// <summary>

/// 평균 제곱 에러 구하기

/// </summary>

/// <param name="trainingValueArray">훈련 값 배열</param>

/// <returns>평균 제곱 에러</returns>

private double GetMeanSquaredError(double[][] trainingValueArray)

{

double summarySquaredError = 0d;

double[] xValueArray = new double[this.inputNodeCount ];

double[] targetValueArray = new double[this.outputNodeCount];

int trainingValueCount = trainingValueArray.Length;

for(int i = 0; i < trainingValueCount; i++)

{

Array.Copy(trainingValueArray[i], xValueArray, this.inputNodeCount);

Array.Copy(trainingValueArray[i], this.inputNodeCount, targetValueArray, 0, this.outputNodeCount);

double[] yValueArray = ComputeOutputNodeValueArray(xValueArray);

for(int j = 0; j < this.outputNodeCount; j++)

{

double error = targetValueArray[j] - yValueArray[j];

summarySquaredError += error * error;

}

return summarySquaredError / trainingValueCount;

}

#endregion

#region 최대 인덱스 구하기 - GetMaximumIndex(vectorArray)

/// <summary>

/// 최대 인덱스 구하기

/// </summary>

/// <param name="vectorArray">벡터 배열</param>

/// <returns>최대 인덱스</returns>

private static int GetMaximumIndex(double[] vectorArray)

{

int maximumIndex = 0;

double maximum = vectorArray[0];

int vectorCount = vectorArray.Length;

for(int i = 0; i < vectorCount; i++)

{

if(vectorArray[i] > maximum)

{

maximum = vectorArray[i];

maximumIndex = i;

}

return maximumIndex;

}

#endregion

}

▶ Program.cs


using System;

namespace TestProject
{
    class Program
    {
        //////////////////////////////////////////////////////////////////////////////////////////////////// Method
        ////////////////////////////////////////////////////////////////////////////////////////// Static
        //////////////////////////////////////////////////////////////////////////////// Private

        #region 프로그램 시작하기 - Main()

        /// <summary>
        /// 프로그램 시작하기
        /// </summary>
        private static void Main()
        {
            Console.Title = "신경망 역전파 알고리즘 사용하기";

            Console.WriteLine();
            Console.WriteLine("신경망 훈련 데모를 시작한다.");
            Console.WriteLine();
            Console.WriteLine("데이터는 유명한 아이리스 꽃 세트이다.");
            Console.WriteLine("꽃받침 길이/너비, 꽃잎 길이/너비에서 종을 예측한다.");
            Console.WriteLine("Iris setosa = 0 0 1, versicolor = 0 1 0, virginica = 1 0 0");
            Console.WriteLine();
            Console.WriteLine("실제 데이터와 유사하다 :");
            Console.WriteLine();
            Console.WriteLine(" 5.1, 3.5, 1.4, 0.2, Iris setosa"    );
            Console.WriteLine(" 7.0, 3.2, 4.7, 1.4, Iris versicolor");
            Console.WriteLine(" 6.3, 3.3, 6.0, 2.5, Iris versinica" );
            Console.WriteLine(" ......");
            Console.WriteLine();

            double[][] sourceValueArray = new double[150][];

            #region 소스 값 배열 데이터를 설정한다.

            sourceValueArray[0] = new double[] { 5.1, 3.5, 1.4, 0.2, 0, 0, 1 };
            sourceValueArray[1] = new double[] { 4.9, 3.0, 1.4, 0.2, 0, 0, 1 }; // Iris setosa = 0 0 1
            sourceValueArray[2] = new double[] { 4.7, 3.2, 1.3, 0.2, 0, 0, 1 }; // Iris versicolor = 0 1 0
            sourceValueArray[3] = new double[] { 4.6, 3.1, 1.5, 0.2, 0, 0, 1 }; // Iris verginica = 1 0 0
            sourceValueArray[4] = new double[] { 5.0, 3.6, 1.4, 0.2, 0, 0, 1 };
            sourceValueArray[5] = new double[] { 5.4, 3.9, 1.7, 0.4, 0, 0, 1 };
            sourceValueArray[6] = new double[] { 4.6, 3.4, 1.4, 0.3, 0, 0, 1 };
            sourceValueArray[7] = new double[] { 5.0, 3.4, 1.5, 0.2, 0, 0, 1 };
            sourceValueArray[8] = new double[] { 4.4, 2.9, 1.4, 0.2, 0, 0, 1 };
            sourceValueArray[9] = new double[] { 4.9, 3.1, 1.5, 0.1, 0, 0, 1 };

            sourceValueArray[10] = new double[] { 5.4, 3.7, 1.5, 0.2, 0, 0, 1 };
            sourceValueArray[11] = new double[] { 4.8, 3.4, 1.6, 0.2, 0, 0, 1 };
            sourceValueArray[12] = new double[] { 4.8, 3.0, 1.4, 0.1, 0, 0, 1 };
            sourceValueArray[13] = new double[] { 4.3, 3.0, 1.1, 0.1, 0, 0, 1 };
            sourceValueArray[14] = new double[] { 5.8, 4.0, 1.2, 0.2, 0, 0, 1 };
            sourceValueArray[15] = new double[] { 5.7, 4.4, 1.5, 0.4, 0, 0, 1 };
            sourceValueArray[16] = new double[] { 5.4, 3.9, 1.3, 0.4, 0, 0, 1 };
            sourceValueArray[17] = new double[] { 5.1, 3.5, 1.4, 0.3, 0, 0, 1 };
            sourceValueArray[18] = new double[] { 5.7, 3.8, 1.7, 0.3, 0, 0, 1 };
            sourceValueArray[19] = new double[] { 5.1, 3.8, 1.5, 0.3, 0, 0, 1 };

            sourceValueArray[20] = new double[] { 5.4, 3.4, 1.7, 0.2, 0, 0, 1 };
            sourceValueArray[21] = new double[] { 5.1, 3.7, 1.5, 0.4, 0, 0, 1 };
            sourceValueArray[22] = new double[] { 4.6, 3.6, 1.0, 0.2, 0, 0, 1 };
            sourceValueArray[23] = new double[] { 5.1, 3.3, 1.7, 0.5, 0, 0, 1 };
            sourceValueArray[24] = new double[] { 4.8, 3.4, 1.9, 0.2, 0, 0, 1 };
            sourceValueArray[25] = new double[] { 5.0, 3.0, 1.6, 0.2, 0, 0, 1 };
            sourceValueArray[26] = new double[] { 5.0, 3.4, 1.6, 0.4, 0, 0, 1 };
            sourceValueArray[27] = new double[] { 5.2, 3.5, 1.5, 0.2, 0, 0, 1 };
            sourceValueArray[28] = new double[] { 5.2, 3.4, 1.4, 0.2, 0, 0, 1 };
            sourceValueArray[29] = new double[] { 4.7, 3.2, 1.6, 0.2, 0, 0, 1 };
            sourceValueArray[30] = new double[] { 4.8, 3.1, 1.6, 0.2, 0, 0, 1 };

            sourceValueArray[31] = new double[] { 5.4, 3.4, 1.5, 0.4, 0, 0, 1 };
            sourceValueArray[32] = new double[] { 5.2, 4.1, 1.5, 0.1, 0, 0, 1 };
            sourceValueArray[33] = new double[] { 5.5, 4.2, 1.4, 0.2, 0, 0, 1 };
            sourceValueArray[34] = new double[] { 4.9, 3.1, 1.5, 0.1, 0, 0, 1 };
            sourceValueArray[35] = new double[] { 5.0, 3.2, 1.2, 0.2, 0, 0, 1 };
            sourceValueArray[36] = new double[] { 5.5, 3.5, 1.3, 0.2, 0, 0, 1 };
            sourceValueArray[37] = new double[] { 4.9, 3.1, 1.5, 0.1, 0, 0, 1 };
            sourceValueArray[38] = new double[] { 4.4, 3.0, 1.3, 0.2, 0, 0, 1 };
            sourceValueArray[39] = new double[] { 5.1, 3.4, 1.5, 0.2, 0, 0, 1 };

            sourceValueArray[40] = new double[] { 5.0, 3.5, 1.3, 0.3, 0, 0, 1 };
            sourceValueArray[41] = new double[] { 4.5, 2.3, 1.3, 0.3, 0, 0, 1 };
            sourceValueArray[42] = new double[] { 4.4, 3.2, 1.3, 0.2, 0, 0, 1 };
            sourceValueArray[43] = new double[] { 5.0, 3.5, 1.6, 0.6, 0, 0, 1 };
            sourceValueArray[44] = new double[] { 5.1, 3.8, 1.9, 0.4, 0, 0, 1 };
            sourceValueArray[45] = new double[] { 4.8, 3.0, 1.4, 0.3, 0, 0, 1 };
            sourceValueArray[46] = new double[] { 5.1, 3.8, 1.6, 0.2, 0, 0, 1 };
            sourceValueArray[47] = new double[] { 4.6, 3.2, 1.4, 0.2, 0, 0, 1 };
            sourceValueArray[48] = new double[] { 5.3, 3.7, 1.5, 0.2, 0, 0, 1 };
            sourceValueArray[49] = new double[] { 5.0, 3.3, 1.4, 0.2, 0, 0, 1 };

            sourceValueArray[50] = new double[] { 7.0, 3.2, 4.7, 1.4, 0, 1, 0 };
            sourceValueArray[51] = new double[] { 6.4, 3.2, 4.5, 1.5, 0, 1, 0 };
            sourceValueArray[52] = new double[] { 6.9, 3.1, 4.9, 1.5, 0, 1, 0 };
            sourceValueArray[53] = new double[] { 5.5, 2.3, 4.0, 1.3, 0, 1, 0 };
            sourceValueArray[54] = new double[] { 6.5, 2.8, 4.6, 1.5, 0, 1, 0 };
            sourceValueArray[55] = new double[] { 5.7, 2.8, 4.5, 1.3, 0, 1, 0 };
            sourceValueArray[56] = new double[] { 6.3, 3.3, 4.7, 1.6, 0, 1, 0 };
            sourceValueArray[57] = new double[] { 4.9, 2.4, 3.3, 1.0, 0, 1, 0 };
            sourceValueArray[58] = new double[] { 6.6, 2.9, 4.6, 1.3, 0, 1, 0 };
            sourceValueArray[59] = new double[] { 5.2, 2.7, 3.9, 1.4, 0, 1, 0 };

            sourceValueArray[60] = new double[] { 5.0, 2.0, 3.5, 1.0, 0, 1, 0 };
            sourceValueArray[61] = new double[] { 5.9, 3.0, 4.2, 1.5, 0, 1, 0 };
            sourceValueArray[62] = new double[] { 6.0, 2.2, 4.0, 1.0, 0, 1, 0 };
            sourceValueArray[63] = new double[] { 6.1, 2.9, 4.7, 1.4, 0, 1, 0 };
            sourceValueArray[64] = new double[] { 5.6, 2.9, 3.6, 1.3, 0, 1, 0 };
            sourceValueArray[65] = new double[] { 6.7, 3.1, 4.4, 1.4, 0, 1, 0 };
            sourceValueArray[66] = new double[] { 5.6, 3.0, 4.5, 1.5, 0, 1, 0 };
            sourceValueArray[67] = new double[] { 5.8, 2.7, 4.1, 1.0, 0, 1, 0 };
            sourceValueArray[68] = new double[] { 6.2, 2.2, 4.5, 1.5, 0, 1, 0 };
            sourceValueArray[69] = new double[] { 5.6, 2.5, 3.9, 1.1, 0, 1, 0 };

            sourceValueArray[70] = new double[] { 5.9, 3.2, 4.8, 1.8, 0, 1, 0 };
            sourceValueArray[71] = new double[] { 6.1, 2.8, 4.0, 1.3, 0, 1, 0 };
            sourceValueArray[72] = new double[] { 6.3, 2.5, 4.9, 1.5, 0, 1, 0 };
            sourceValueArray[73] = new double[] { 6.1, 2.8, 4.7, 1.2, 0, 1, 0 };
            sourceValueArray[74] = new double[] { 6.4, 2.9, 4.3, 1.3, 0, 1, 0 };
            sourceValueArray[75] = new double[] { 6.6, 3.0, 4.4, 1.4, 0, 1, 0 };
            sourceValueArray[76] = new double[] { 6.8, 2.8, 4.8, 1.4, 0, 1, 0 };
            sourceValueArray[77] = new double[] { 6.7, 3.0, 5.0, 1.7, 0, 1, 0 };
            sourceValueArray[78] = new double[] { 6.0, 2.9, 4.5, 1.5, 0, 1, 0 };
            sourceValueArray[79] = new double[] { 5.7, 2.6, 3.5, 1.0, 0, 1, 0 };

            sourceValueArray[80] = new double[] { 5.5, 2.4, 3.8, 1.1, 0, 1, 0 };
            sourceValueArray[81] = new double[] { 5.5, 2.4, 3.7, 1.0, 0, 1, 0 };
            sourceValueArray[82] = new double[] { 5.8, 2.7, 3.9, 1.2, 0, 1, 0 };
            sourceValueArray[83] = new double[] { 6.0, 2.7, 5.1, 1.6, 0, 1, 0 };
            sourceValueArray[84] = new double[] { 5.4, 3.0, 4.5, 1.5, 0, 1, 0 };
            sourceValueArray[85] = new double[] { 6.0, 3.4, 4.5, 1.6, 0, 1, 0 };
            sourceValueArray[86] = new double[] { 6.7, 3.1, 4.7, 1.5, 0, 1, 0 };
            sourceValueArray[87] = new double[] { 6.3, 2.3, 4.4, 1.3, 0, 1, 0 };
            sourceValueArray[88] = new double[] { 5.6, 3.0, 4.1, 1.3, 0, 1, 0 };
            sourceValueArray[89] = new double[] { 5.5, 2.5, 4.0, 1.3, 0, 1, 0 };

            sourceValueArray[90] = new double[] { 5.5, 2.6, 4.4, 1.2, 0, 1, 0 };
            sourceValueArray[91] = new double[] { 6.1, 3.0, 4.6, 1.4, 0, 1, 0 };
            sourceValueArray[92] = new double[] { 5.8, 2.6, 4.0, 1.2, 0, 1, 0 };
            sourceValueArray[93] = new double[] { 5.0, 2.3, 3.3, 1.0, 0, 1, 0 };
            sourceValueArray[94] = new double[] { 5.6, 2.7, 4.2, 1.3, 0, 1, 0 };
            sourceValueArray[95] = new double[] { 5.7, 3.0, 4.2, 1.2, 0, 1, 0 };
            sourceValueArray[96] = new double[] { 5.7, 2.9, 4.2, 1.3, 0, 1, 0 };
            sourceValueArray[97] = new double[] { 6.2, 2.9, 4.3, 1.3, 0, 1, 0 };
            sourceValueArray[98] = new double[] { 5.1, 2.5, 3.0, 1.1, 0, 1, 0 };
            sourceValueArray[99] = new double[] { 5.7, 2.8, 4.1, 1.3, 0, 1, 0 };

            sourceValueArray[100] = new double[] { 6.3, 3.3, 6.0, 2.5, 1, 0, 0 };
            sourceValueArray[101] = new double[] { 5.8, 2.7, 5.1, 1.9, 1, 0, 0 };
            sourceValueArray[102] = new double[] { 7.1, 3.0, 5.9, 2.1, 1, 0, 0 };
            sourceValueArray[103] = new double[] { 6.3, 2.9, 5.6, 1.8, 1, 0, 0 };
            sourceValueArray[104] = new double[] { 6.5, 3.0, 5.8, 2.2, 1, 0, 0 };
            sourceValueArray[105] = new double[] { 7.6, 3.0, 6.6, 2.1, 1, 0, 0 };
            sourceValueArray[106] = new double[] { 4.9, 2.5, 4.5, 1.7, 1, 0, 0 };
            sourceValueArray[107] = new double[] { 7.3, 2.9, 6.3, 1.8, 1, 0, 0 };
            sourceValueArray[108] = new double[] { 6.7, 2.5, 5.8, 1.8, 1, 0, 0 };
            sourceValueArray[109] = new double[] { 7.2, 3.6, 6.1, 2.5, 1, 0, 0 };

            sourceValueArray[110] = new double[] { 6.5, 3.2, 5.1, 2.0, 1, 0, 0 };
            sourceValueArray[111] = new double[] { 6.4, 2.7, 5.3, 1.9, 1, 0, 0 };
            sourceValueArray[112] = new double[] { 6.8, 3.0, 5.5, 2.1, 1, 0, 0 };
            sourceValueArray[113] = new double[] { 5.7, 2.5, 5.0, 2.0, 1, 0, 0 };
            sourceValueArray[114] = new double[] { 5.8, 2.8, 5.1, 2.4, 1, 0, 0 };
            sourceValueArray[115] = new double[] { 6.4, 3.2, 5.3, 2.3, 1, 0, 0 };
            sourceValueArray[116] = new double[] { 6.5, 3.0, 5.5, 1.8, 1, 0, 0 };
            sourceValueArray[117] = new double[] { 7.7, 3.8, 6.7, 2.2, 1, 0, 0 };
            sourceValueArray[118] = new double[] { 7.7, 2.6, 6.9, 2.3, 1, 0, 0 };
            sourceValueArray[119] = new double[] { 6.0, 2.2, 5.0, 1.5, 1, 0, 0 };

            sourceValueArray[120] = new double[] { 6.9, 3.2, 5.7, 2.3, 1, 0, 0 };
            sourceValueArray[121] = new double[] { 5.6, 2.8, 4.9, 2.0, 1, 0, 0 };
            sourceValueArray[122] = new double[] { 7.7, 2.8, 6.7, 2.0, 1, 0, 0 };
            sourceValueArray[123] = new double[] { 6.3, 2.7, 4.9, 1.8, 1, 0, 0 };
            sourceValueArray[124] = new double[] { 6.7, 3.3, 5.7, 2.1, 1, 0, 0 };
            sourceValueArray[125] = new double[] { 7.2, 3.2, 6.0, 1.8, 1, 0, 0 };
            sourceValueArray[126] = new double[] { 6.2, 2.8, 4.8, 1.8, 1, 0, 0 };
            sourceValueArray[127] = new double[] { 6.1, 3.0, 4.9, 1.8, 1, 0, 0 };
            sourceValueArray[128] = new double[] { 6.4, 2.8, 5.6, 2.1, 1, 0, 0 };
            sourceValueArray[129] = new double[] { 7.2, 3.0, 5.8, 1.6, 1, 0, 0 };

            sourceValueArray[130] = new double[] { 7.4, 2.8, 6.1, 1.9, 1, 0, 0 };
            sourceValueArray[131] = new double[] { 7.9, 3.8, 6.4, 2.0, 1, 0, 0 };
            sourceValueArray[132] = new double[] { 6.4, 2.8, 5.6, 2.2, 1, 0, 0 };
            sourceValueArray[133] = new double[] { 6.3, 2.8, 5.1, 1.5, 1, 0, 0 };
            sourceValueArray[134] = new double[] { 6.1, 2.6, 5.6, 1.4, 1, 0, 0 };
            sourceValueArray[135] = new double[] { 7.7, 3.0, 6.1, 2.3, 1, 0, 0 };
            sourceValueArray[136] = new double[] { 6.3, 3.4, 5.6, 2.4, 1, 0, 0 };
            sourceValueArray[137] = new double[] { 6.4, 3.1, 5.5, 1.8, 1, 0, 0 };
            sourceValueArray[138] = new double[] { 6.0, 3.0, 4.8, 1.8, 1, 0, 0 };
            sourceValueArray[139] = new double[] { 6.9, 3.1, 5.4, 2.1, 1, 0, 0 };

            sourceValueArray[140] = new double[] { 6.7, 3.1, 5.6, 2.4, 1, 0, 0 };
            sourceValueArray[141] = new double[] { 6.9, 3.1, 5.1, 2.3, 1, 0, 0 };
            sourceValueArray[142] = new double[] { 5.8, 2.7, 5.1, 1.9, 1, 0, 0 };
            sourceValueArray[143] = new double[] { 6.8, 3.2, 5.9, 2.3, 1, 0, 0 };
            sourceValueArray[144] = new double[] { 6.7, 3.3, 5.7, 2.5, 1, 0, 0 };
            sourceValueArray[145] = new double[] { 6.7, 3.0, 5.2, 2.3, 1, 0, 0 };
            sourceValueArray[146] = new double[] { 6.3, 2.5, 5.0, 1.9, 1, 0, 0 };
            sourceValueArray[147] = new double[] { 6.5, 3.0, 5.2, 2.0, 1, 0, 0 };
            sourceValueArray[148] = new double[] { 6.2, 3.4, 5.4, 2.3, 1, 0, 0 };
            sourceValueArray[149] = new double[] { 5.9, 3.0, 5.1, 1.8, 1, 0, 0 };

            #endregion

            Console.WriteLine("150개 항목 데이터 세트의 첫번째 6개 행들 :");
            Console.WriteLine();

            DisplayMatrixArray(sourceValueArray, 6, 1, true);

            Console.WriteLine("80% 훈련 데이터 매트릭스와 20% 테스트 데이터 매트릭스 생성");

            double[][] traingingValueArray = null;
            double[][] testValueArray      = null;

            PrepareData(sourceValueArray, 72, out traingingValueArray, out testValueArray);

            Console.WriteLine();
            Console.WriteLine("훈련 데이터의 첫번째 3개 행들 :");
            Console.WriteLine();

            DisplayMatrixArray(traingingValueArray, 3, 1, true);

            Console.WriteLine("테스트 데이터의 첫번째 3개 행들 :");
            Console.WriteLine();

            DisplayMatrixArray(testValueArray, 3, 1, true);

            Console.WriteLine("4개 입력 노드, 7개 은닉 노드, 3개 출력 노드를 갖는 신경망 생성");
            Console.WriteLine("입력-은닉층을 위한 하드 코딩된 하이퍼 탄젠트 함수와 은닉-출력층 활성화를 위한 소프트 맥스 함수");

            int inputNodeCount  = 4;
            int hiddenModeCount = 7;
            int outputNodeCount = 3;

            NeuralNetwork neuralNetwork = new NeuralNetwork(inputNodeCount, hiddenModeCount, outputNodeCount);

            int    maximumEpochCount = 1000;
            double learningRate      = 0.05d;
            double momentum          = 0.01d;

            Console.WriteLine("최대 시대 카운트 = " + maximumEpochCount + ", 학습률 = " + learningRate + ", 모멘텀 = " + momentum);

            Console.WriteLine("훈련은 하드 코딩된 평균 제곱 오류가 0.040보다 작은 경우 중단하는 조건을 갖는다.");

            Console.WriteLine();
            Console.WriteLine("증분 역전파를 사용해 훈련 시작하기");
            Console.WriteLine();

            neuralNetwork.Train(traingingValueArray, maximumEpochCount, learningRate, momentum);

            Console.WriteLine("훈련 완료");
            Console.WriteLine();

            double[] weightArray = neuralNetwork.GetWeightArray();

            Console.WriteLine("최종 신경망 가중치와 바이어스 값들 : ");

            DisplayVectorArray(weightArray, 10, 3, true);

            double trainingAccuracy = neuralNetwork.GetAccuracy(traingingValueArray);
            double testAccuracy     = neuralNetwork.GetAccuracy(testValueArray     );

            Console.WriteLine();
            Console.WriteLine("훈련 데이터 정확도   = " + trainingAccuracy.ToString("F4"));
            Console.WriteLine("테스트 데이터 정확도 = " + testAccuracy.ToString("F4"));
            Console.WriteLine();

            Console.WriteLine("입력 데이터 : ");

            double[] inputArray = new double[] { 5.1, 3.5, 1.4, 0.2 };

            DisplayVectorArray(inputArray, 4, 3, true);

            double[] resultArray = neuralNetwork.ComputeOutputNodeValueArray(inputArray);

            Console.WriteLine();
            Console.WriteLine("출력 데이터 : ");

            DisplayVectorArray(resultArray, 3, 3, true);

            Console.WriteLine();
            Console.WriteLine("신경망 훈련 데모 종료");

            Console.ReadLine();
        }

        #endregion

        #region 데이터 준비하기 - PrepareData(sourceValueArray, seed, trainingValueArray, testValueArray)

        /// <summary>
        /// 데이터 준비하기
        /// </summary>
        /// <param name="sourceValueArray">소스 값 배열</param>
        /// <param name="seed">시드</param>
        /// <param name="trainingValueArray">훈련 값 배열</param>
        /// <param name="testValueArray">테스트 값 배열</param>
        private static void PrepareData
        (
            double[][]     sourceValueArray,
            int            seed,
            out double[][] trainingValueArray,
            out double[][] testValueArray
        )
        {
            Random random = new Random(seed);

            int rowCount         = sourceValueArray.Length;
            int columnCount      = sourceValueArray[0].Length;
            int trainingRowCount = (int)(rowCount * 0.80);
            int testRowCount     = rowCount - trainingRowCount;

            trainingValueArray = new double[trainingRowCount][];
            testValueArray     = new double[testRowCount][];

            double[][] copyValueArray = new double[sourceValueArray.Length][];

            for(int i = 0; i < copyValueArray.Length; i++)
            {
                copyValueArray[i] = sourceValueArray[i];
            }

            for(int i = 0; i < copyValueArray.Length; i++)
            {
                int randomIndex = random.Next(i, copyValueArray.Length);

                double[] temporaryArray = copyValueArray[randomIndex];

                copyValueArray[randomIndex] = copyValueArray[i];
                copyValueArray[i          ] = temporaryArray;
            }

            for(int i = 0; i < trainingRowCount; i++)
            {
                trainingValueArray[i] = new double[columnCount];

                for(int j = 0; j < columnCount; j++)
                {
                    trainingValueArray[i][j] = copyValueArray[i][j];
                }
            }

            for(int i = 0; i < testRowCount; i++)
            {
                testValueArray[i] = new double[columnCount];

                for(int j = 0; j < columnCount; j++)
                {
                    testValueArray[i][j] = copyValueArray[i + trainingRowCount][j];
                }
            }
        }

        #endregion
        #region 벡터 배열 출력하기 - DisplayVectorArray(sourceVectorArray, valueCountPerRow, decimalCount, newLine)

        /// <summary>
        /// 벡터 배열 출력하기
        /// </summary>
        /// <param name="sourceVectorArray">소스 벡터 배열</param>
        /// <param name="valueCountPerRow">행당 값 카운트</param>
        /// <param name="decimalCount">소수점 카운트</param>
        /// <param name="newLine">개행 여부</param>
        private static void DisplayVectorArray(double[] sourceVectorArray, int valueCountPerRow, int decimalCount, bool newLine)
        {
            int sourceVectorCount = sourceVectorArray.Length;

            for(int i = 0; i < sourceVectorCount; i++)
            {
                if(i % valueCountPerRow == 0)
                {
                    Console.WriteLine("");
                }

                Console.Write(sourceVectorArray[i].ToString("F" + decimalCount).PadLeft(decimalCount + 4) + " ");
            }

            if (newLine == true)
            {
                Console.WriteLine("");
            }
        }

        #endregion
        #region 매트릭스 배열 출력하기 - DisplayMatrixArray(sourceMatrixArray, rowCount, decimalCount, newLine)

        /// <summary>
        /// 매트릭스 배열 출력하기
        /// </summary>
        /// <param name="sourceMatrixArray">소스 매트릭스 배열</param>
        /// <param name="rowCount">행 카운트</param>
        /// <param name="decimalCount">소수점 카운트</param>
        /// <param name="newLine">개행 여부</param>
        private static void DisplayMatrixArray(double[][] sourceMatrixArray, int rowCount, int decimalCount, bool newLine)
        {
            for(int i = 0; i < rowCount; i++)
            {
                Console.Write(i.ToString().PadLeft(3) + " : ");

                for(int j = 0; j < sourceMatrixArray[i].Length; j++)
                {
                    if(sourceMatrixArray[i][j] >= 0d)
                    {
                        Console.Write(" ");
                    }
                    else
                    {
                        Console.Write("-");
                    }

                    Console.Write(Math.Abs(sourceMatrixArray[i][j]).ToString("F" + decimalCount) + " ");
                }

                Console.WriteLine("");
            }

            if(newLine == true)
            {
                Console.WriteLine("");
            }
        }

        #endregion
    }
}

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

using System;

namespace TestProject

{

class Program

{

//////////////////////////////////////////////////////////////////////////////////////////////////// Method

////////////////////////////////////////////////////////////////////////////////////////// Static

//////////////////////////////////////////////////////////////////////////////// Private

#region 프로그램 시작하기 - Main()

/// <summary>

/// 프로그램 시작하기

/// </summary>

private static void Main()

{

Console.Title = "신경망 역전파 알고리즘 사용하기";

Console.WriteLine();

Console.WriteLine("신경망 훈련 데모를 시작한다.");

Console.WriteLine();

Console.WriteLine("데이터는 유명한 아이리스 꽃 세트이다.");

Console.WriteLine("꽃받침 길이/너비, 꽃잎 길이/너비에서 종을 예측한다.");

Console.WriteLine("Iris setosa = 0 0 1, versicolor = 0 1 0, virginica = 1 0 0");

Console.WriteLine();

Console.WriteLine("실제 데이터와 유사하다 :");

Console.WriteLine();

Console.WriteLine(" 5.1, 3.5, 1.4, 0.2, Iris setosa" );

Console.WriteLine(" 7.0, 3.2, 4.7, 1.4, Iris versicolor");

Console.WriteLine(" 6.3, 3.3, 6.0, 2.5, Iris versinica" );

Console.WriteLine(" ......");

Console.WriteLine();

double[][] sourceValueArray = new double[150][];

#region 소스 값 배열 데이터를 설정한다.

sourceValueArray[0] = new double[] { 5.1, 3.5, 1.4, 0.2, 0, 0, 1 };

sourceValueArray[1] = new double[] { 4.9, 3.0, 1.4, 0.2, 0, 0, 1 }; // Iris setosa = 0 0 1

sourceValueArray[2] = new double[] { 4.7, 3.2, 1.3, 0.2, 0, 0, 1 }; // Iris versicolor = 0 1 0

sourceValueArray[3] = new double[] { 4.6, 3.1, 1.5, 0.2, 0, 0, 1 }; // Iris verginica = 1 0 0

sourceValueArray[4] = new double[] { 5.0, 3.6, 1.4, 0.2, 0, 0, 1 };

sourceValueArray[5] = new double[] { 5.4, 3.9, 1.7, 0.4, 0, 0, 1 };

sourceValueArray[6] = new double[] { 4.6, 3.4, 1.4, 0.3, 0, 0, 1 };

sourceValueArray[7] = new double[] { 5.0, 3.4, 1.5, 0.2, 0, 0, 1 };

sourceValueArray[8] = new double[] { 4.4, 2.9, 1.4, 0.2, 0, 0, 1 };

sourceValueArray[9] = new double[] { 4.9, 3.1, 1.5, 0.1, 0, 0, 1 };

sourceValueArray[10] = new double[] { 5.4, 3.7, 1.5, 0.2, 0, 0, 1 };

sourceValueArray[11] = new double[] { 4.8, 3.4, 1.6, 0.2, 0, 0, 1 };

sourceValueArray[12] = new double[] { 4.8, 3.0, 1.4, 0.1, 0, 0, 1 };

sourceValueArray[13] = new double[] { 4.3, 3.0, 1.1, 0.1, 0, 0, 1 };

sourceValueArray[14] = new double[] { 5.8, 4.0, 1.2, 0.2, 0, 0, 1 };

sourceValueArray[15] = new double[] { 5.7, 4.4, 1.5, 0.4, 0, 0, 1 };

sourceValueArray[16] = new double[] { 5.4, 3.9, 1.3, 0.4, 0, 0, 1 };

sourceValueArray[17] = new double[] { 5.1, 3.5, 1.4, 0.3, 0, 0, 1 };

sourceValueArray[18] = new double[] { 5.7, 3.8, 1.7, 0.3, 0, 0, 1 };

sourceValueArray[19] = new double[] { 5.1, 3.8, 1.5, 0.3, 0, 0, 1 };

sourceValueArray[20] = new double[] { 5.4, 3.4, 1.7, 0.2, 0, 0, 1 };

sourceValueArray[21] = new double[] { 5.1, 3.7, 1.5, 0.4, 0, 0, 1 };

sourceValueArray[22] = new double[] { 4.6, 3.6, 1.0, 0.2, 0, 0, 1 };

sourceValueArray[23] = new double[] { 5.1, 3.3, 1.7, 0.5, 0, 0, 1 };

sourceValueArray[24] = new double[] { 4.8, 3.4, 1.9, 0.2, 0, 0, 1 };

sourceValueArray[25] = new double[] { 5.0, 3.0, 1.6, 0.2, 0, 0, 1 };

sourceValueArray[26] = new double[] { 5.0, 3.4, 1.6, 0.4, 0, 0, 1 };

sourceValueArray[27] = new double[] { 5.2, 3.5, 1.5, 0.2, 0, 0, 1 };

sourceValueArray[28] = new double[] { 5.2, 3.4, 1.4, 0.2, 0, 0, 1 };

sourceValueArray[29] = new double[] { 4.7, 3.2, 1.6, 0.2, 0, 0, 1 };

sourceValueArray[30] = new double[] { 4.8, 3.1, 1.6, 0.2, 0, 0, 1 };

sourceValueArray[31] = new double[] { 5.4, 3.4, 1.5, 0.4, 0, 0, 1 };

sourceValueArray[32] = new double[] { 5.2, 4.1, 1.5, 0.1, 0, 0, 1 };

sourceValueArray[33] = new double[] { 5.5, 4.2, 1.4, 0.2, 0, 0, 1 };

sourceValueArray[34] = new double[] { 4.9, 3.1, 1.5, 0.1, 0, 0, 1 };

sourceValueArray[35] = new double[] { 5.0, 3.2, 1.2, 0.2, 0, 0, 1 };

sourceValueArray[36] = new double[] { 5.5, 3.5, 1.3, 0.2, 0, 0, 1 };

sourceValueArray[37] = new double[] { 4.9, 3.1, 1.5, 0.1, 0, 0, 1 };

sourceValueArray[38] = new double[] { 4.4, 3.0, 1.3, 0.2, 0, 0, 1 };

sourceValueArray[39] = new double[] { 5.1, 3.4, 1.5, 0.2, 0, 0, 1 };

sourceValueArray[40] = new double[] { 5.0, 3.5, 1.3, 0.3, 0, 0, 1 };

sourceValueArray[41] = new double[] { 4.5, 2.3, 1.3, 0.3, 0, 0, 1 };

sourceValueArray[42] = new double[] { 4.4, 3.2, 1.3, 0.2, 0, 0, 1 };

sourceValueArray[43] = new double[] { 5.0, 3.5, 1.6, 0.6, 0, 0, 1 };

sourceValueArray[44] = new double[] { 5.1, 3.8, 1.9, 0.4, 0, 0, 1 };

sourceValueArray[45] = new double[] { 4.8, 3.0, 1.4, 0.3, 0, 0, 1 };

sourceValueArray[46] = new double[] { 5.1, 3.8, 1.6, 0.2, 0, 0, 1 };

sourceValueArray[47] = new double[] { 4.6, 3.2, 1.4, 0.2, 0, 0, 1 };

sourceValueArray[48] = new double[] { 5.3, 3.7, 1.5, 0.2, 0, 0, 1 };

sourceValueArray[49] = new double[] { 5.0, 3.3, 1.4, 0.2, 0, 0, 1 };

sourceValueArray[50] = new double[] { 7.0, 3.2, 4.7, 1.4, 0, 1, 0 };

sourceValueArray[51] = new double[] { 6.4, 3.2, 4.5, 1.5, 0, 1, 0 };

sourceValueArray[52] = new double[] { 6.9, 3.1, 4.9, 1.5, 0, 1, 0 };

sourceValueArray[53] = new double[] { 5.5, 2.3, 4.0, 1.3, 0, 1, 0 };

sourceValueArray[54] = new double[] { 6.5, 2.8, 4.6, 1.5, 0, 1, 0 };

sourceValueArray[55] = new double[] { 5.7, 2.8, 4.5, 1.3, 0, 1, 0 };

sourceValueArray[56] = new double[] { 6.3, 3.3, 4.7, 1.6, 0, 1, 0 };

sourceValueArray[57] = new double[] { 4.9, 2.4, 3.3, 1.0, 0, 1, 0 };

sourceValueArray[58] = new double[] { 6.6, 2.9, 4.6, 1.3, 0, 1, 0 };

sourceValueArray[59] = new double[] { 5.2, 2.7, 3.9, 1.4, 0, 1, 0 };

sourceValueArray[60] = new double[] { 5.0, 2.0, 3.5, 1.0, 0, 1, 0 };

sourceValueArray[61] = new double[] { 5.9, 3.0, 4.2, 1.5, 0, 1, 0 };

sourceValueArray[62] = new double[] { 6.0, 2.2, 4.0, 1.0, 0, 1, 0 };

sourceValueArray[63] = new double[] { 6.1, 2.9, 4.7, 1.4, 0, 1, 0 };

sourceValueArray[64] = new double[] { 5.6, 2.9, 3.6, 1.3, 0, 1, 0 };

sourceValueArray[65] = new double[] { 6.7, 3.1, 4.4, 1.4, 0, 1, 0 };

sourceValueArray[66] = new double[] { 5.6, 3.0, 4.5, 1.5, 0, 1, 0 };

sourceValueArray[67] = new double[] { 5.8, 2.7, 4.1, 1.0, 0, 1, 0 };

sourceValueArray[68] = new double[] { 6.2, 2.2, 4.5, 1.5, 0, 1, 0 };

sourceValueArray[69] = new double[] { 5.6, 2.5, 3.9, 1.1, 0, 1, 0 };

sourceValueArray[70] = new double[] { 5.9, 3.2, 4.8, 1.8, 0, 1, 0 };

sourceValueArray[71] = new double[] { 6.1, 2.8, 4.0, 1.3, 0, 1, 0 };

sourceValueArray[72] = new double[] { 6.3, 2.5, 4.9, 1.5, 0, 1, 0 };

sourceValueArray[73] = new double[] { 6.1, 2.8, 4.7, 1.2, 0, 1, 0 };

sourceValueArray[74] = new double[] { 6.4, 2.9, 4.3, 1.3, 0, 1, 0 };

sourceValueArray[75] = new double[] { 6.6, 3.0, 4.4, 1.4, 0, 1, 0 };

sourceValueArray[76] = new double[] { 6.8, 2.8, 4.8, 1.4, 0, 1, 0 };

sourceValueArray[77] = new double[] { 6.7, 3.0, 5.0, 1.7, 0, 1, 0 };

sourceValueArray[78] = new double[] { 6.0, 2.9, 4.5, 1.5, 0, 1, 0 };

sourceValueArray[79] = new double[] { 5.7, 2.6, 3.5, 1.0, 0, 1, 0 };

sourceValueArray[80] = new double[] { 5.5, 2.4, 3.8, 1.1, 0, 1, 0 };

sourceValueArray[81] = new double[] { 5.5, 2.4, 3.7, 1.0, 0, 1, 0 };

sourceValueArray[82] = new double[] { 5.8, 2.7, 3.9, 1.2, 0, 1, 0 };

sourceValueArray[83] = new double[] { 6.0, 2.7, 5.1, 1.6, 0, 1, 0 };

sourceValueArray[84] = new double[] { 5.4, 3.0, 4.5, 1.5, 0, 1, 0 };

sourceValueArray[85] = new double[] { 6.0, 3.4, 4.5, 1.6, 0, 1, 0 };

sourceValueArray[86] = new double[] { 6.7, 3.1, 4.7, 1.5, 0, 1, 0 };

sourceValueArray[87] = new double[] { 6.3, 2.3, 4.4, 1.3, 0, 1, 0 };

sourceValueArray[88] = new double[] { 5.6, 3.0, 4.1, 1.3, 0, 1, 0 };

sourceValueArray[89] = new double[] { 5.5, 2.5, 4.0, 1.3, 0, 1, 0 };

sourceValueArray[90] = new double[] { 5.5, 2.6, 4.4, 1.2, 0, 1, 0 };

sourceValueArray[91] = new double[] { 6.1, 3.0, 4.6, 1.4, 0, 1, 0 };

sourceValueArray[92] = new double[] { 5.8, 2.6, 4.0, 1.2, 0, 1, 0 };

sourceValueArray[93] = new double[] { 5.0, 2.3, 3.3, 1.0, 0, 1, 0 };

sourceValueArray[94] = new double[] { 5.6, 2.7, 4.2, 1.3, 0, 1, 0 };

sourceValueArray[95] = new double[] { 5.7, 3.0, 4.2, 1.2, 0, 1, 0 };

sourceValueArray[96] = new double[] { 5.7, 2.9, 4.2, 1.3, 0, 1, 0 };

sourceValueArray[97] = new double[] { 6.2, 2.9, 4.3, 1.3, 0, 1, 0 };

sourceValueArray[98] = new double[] { 5.1, 2.5, 3.0, 1.1, 0, 1, 0 };

sourceValueArray[99] = new double[] { 5.7, 2.8, 4.1, 1.3, 0, 1, 0 };

sourceValueArray[100] = new double[] { 6.3, 3.3, 6.0, 2.5, 1, 0, 0 };

sourceValueArray[101] = new double[] { 5.8, 2.7, 5.1, 1.9, 1, 0, 0 };

sourceValueArray[102] = new double[] { 7.1, 3.0, 5.9, 2.1, 1, 0, 0 };

sourceValueArray[103] = new double[] { 6.3, 2.9, 5.6, 1.8, 1, 0, 0 };

sourceValueArray[104] = new double[] { 6.5, 3.0, 5.8, 2.2, 1, 0, 0 };

sourceValueArray[105] = new double[] { 7.6, 3.0, 6.6, 2.1, 1, 0, 0 };

sourceValueArray[106] = new double[] { 4.9, 2.5, 4.5, 1.7, 1, 0, 0 };

sourceValueArray[107] = new double[] { 7.3, 2.9, 6.3, 1.8, 1, 0, 0 };

sourceValueArray[108] = new double[] { 6.7, 2.5, 5.8, 1.8, 1, 0, 0 };

sourceValueArray[109] = new double[] { 7.2, 3.6, 6.1, 2.5, 1, 0, 0 };

sourceValueArray[110] = new double[] { 6.5, 3.2, 5.1, 2.0, 1, 0, 0 };

sourceValueArray[111] = new double[] { 6.4, 2.7, 5.3, 1.9, 1, 0, 0 };

sourceValueArray[112] = new double[] { 6.8, 3.0, 5.5, 2.1, 1, 0, 0 };

sourceValueArray[113] = new double[] { 5.7, 2.5, 5.0, 2.0, 1, 0, 0 };

sourceValueArray[114] = new double[] { 5.8, 2.8, 5.1, 2.4, 1, 0, 0 };

sourceValueArray[115] = new double[] { 6.4, 3.2, 5.3, 2.3, 1, 0, 0 };

sourceValueArray[116] = new double[] { 6.5, 3.0, 5.5, 1.8, 1, 0, 0 };

sourceValueArray[117] = new double[] { 7.7, 3.8, 6.7, 2.2, 1, 0, 0 };

sourceValueArray[118] = new double[] { 7.7, 2.6, 6.9, 2.3, 1, 0, 0 };

sourceValueArray[119] = new double[] { 6.0, 2.2, 5.0, 1.5, 1, 0, 0 };

sourceValueArray[120] = new double[] { 6.9, 3.2, 5.7, 2.3, 1, 0, 0 };

sourceValueArray[121] = new double[] { 5.6, 2.8, 4.9, 2.0, 1, 0, 0 };

sourceValueArray[122] = new double[] { 7.7, 2.8, 6.7, 2.0, 1, 0, 0 };

sourceValueArray[123] = new double[] { 6.3, 2.7, 4.9, 1.8, 1, 0, 0 };

sourceValueArray[124] = new double[] { 6.7, 3.3, 5.7, 2.1, 1, 0, 0 };

sourceValueArray[125] = new double[] { 7.2, 3.2, 6.0, 1.8, 1, 0, 0 };

sourceValueArray[126] = new double[] { 6.2, 2.8, 4.8, 1.8, 1, 0, 0 };

sourceValueArray[127] = new double[] { 6.1, 3.0, 4.9, 1.8, 1, 0, 0 };

sourceValueArray[128] = new double[] { 6.4, 2.8, 5.6, 2.1, 1, 0, 0 };

sourceValueArray[129] = new double[] { 7.2, 3.0, 5.8, 1.6, 1, 0, 0 };

sourceValueArray[130] = new double[] { 7.4, 2.8, 6.1, 1.9, 1, 0, 0 };

sourceValueArray[131] = new double[] { 7.9, 3.8, 6.4, 2.0, 1, 0, 0 };

sourceValueArray[132] = new double[] { 6.4, 2.8, 5.6, 2.2, 1, 0, 0 };

sourceValueArray[133] = new double[] { 6.3, 2.8, 5.1, 1.5, 1, 0, 0 };

sourceValueArray[134] = new double[] { 6.1, 2.6, 5.6, 1.4, 1, 0, 0 };

sourceValueArray[135] = new double[] { 7.7, 3.0, 6.1, 2.3, 1, 0, 0 };

sourceValueArray[136] = new double[] { 6.3, 3.4, 5.6, 2.4, 1, 0, 0 };

sourceValueArray[137] = new double[] { 6.4, 3.1, 5.5, 1.8, 1, 0, 0 };

sourceValueArray[138] = new double[] { 6.0, 3.0, 4.8, 1.8, 1, 0, 0 };

sourceValueArray[139] = new double[] { 6.9, 3.1, 5.4, 2.1, 1, 0, 0 };

sourceValueArray[140] = new double[] { 6.7, 3.1, 5.6, 2.4, 1, 0, 0 };

sourceValueArray[141] = new double[] { 6.9, 3.1, 5.1, 2.3, 1, 0, 0 };

sourceValueArray[142] = new double[] { 5.8, 2.7, 5.1, 1.9, 1, 0, 0 };

sourceValueArray[143] = new double[] { 6.8, 3.2, 5.9, 2.3, 1, 0, 0 };

sourceValueArray[144] = new double[] { 6.7, 3.3, 5.7, 2.5, 1, 0, 0 };

sourceValueArray[145] = new double[] { 6.7, 3.0, 5.2, 2.3, 1, 0, 0 };

sourceValueArray[146] = new double[] { 6.3, 2.5, 5.0, 1.9, 1, 0, 0 };

sourceValueArray[147] = new double[] { 6.5, 3.0, 5.2, 2.0, 1, 0, 0 };

sourceValueArray[148] = new double[] { 6.2, 3.4, 5.4, 2.3, 1, 0, 0 };

sourceValueArray[149] = new double[] { 5.9, 3.0, 5.1, 1.8, 1, 0, 0 };

#endregion

Console.WriteLine("150개 항목 데이터 세트의 첫번째 6개 행들 :");

Console.WriteLine();

DisplayMatrixArray(sourceValueArray, 6, 1, true);

Console.WriteLine("80% 훈련 데이터 매트릭스와 20% 테스트 데이터 매트릭스 생성");

double[][] traingingValueArray = null;

double[][] testValueArray = null;

PrepareData(sourceValueArray, 72, out traingingValueArray, out testValueArray);

Console.WriteLine();

Console.WriteLine("훈련 데이터의 첫번째 3개 행들 :");

Console.WriteLine();

DisplayMatrixArray(traingingValueArray, 3, 1, true);

Console.WriteLine("테스트 데이터의 첫번째 3개 행들 :");

Console.WriteLine();

DisplayMatrixArray(testValueArray, 3, 1, true);

Console.WriteLine("4개 입력 노드, 7개 은닉 노드, 3개 출력 노드를 갖는 신경망 생성");

Console.WriteLine("입력-은닉층을 위한 하드 코딩된 하이퍼 탄젠트 함수와 은닉-출력층 활성화를 위한 소프트 맥스 함수");

int inputNodeCount = 4;

int hiddenModeCount = 7;

int outputNodeCount = 3;

NeuralNetwork neuralNetwork = new NeuralNetwork(inputNodeCount, hiddenModeCount, outputNodeCount);

int maximumEpochCount = 1000;

double learningRate = 0.05d;

double momentum = 0.01d;

Console.WriteLine("최대 시대 카운트 = " + maximumEpochCount + ", 학습률 = " + learningRate + ", 모멘텀 = " + momentum);

Console.WriteLine("훈련은 하드 코딩된 평균 제곱 오류가 0.040보다 작은 경우 중단하는 조건을 갖는다.");

Console.WriteLine();

Console.WriteLine("증분 역전파를 사용해 훈련 시작하기");

Console.WriteLine();

neuralNetwork.Train(traingingValueArray, maximumEpochCount, learningRate, momentum);

Console.WriteLine("훈련 완료");

Console.WriteLine();

double[] weightArray = neuralNetwork.GetWeightArray();

Console.WriteLine("최종 신경망 가중치와 바이어스 값들 : ");

DisplayVectorArray(weightArray, 10, 3, true);

double trainingAccuracy = neuralNetwork.GetAccuracy(traingingValueArray);

double testAccuracy = neuralNetwork.GetAccuracy(testValueArray );

Console.WriteLine();

Console.WriteLine("훈련 데이터 정확도 = " + trainingAccuracy.ToString("F4"));

Console.WriteLine("테스트 데이터 정확도 = " + testAccuracy.ToString("F4"));

Console.WriteLine();

Console.WriteLine("입력 데이터 : ");

double[] inputArray = new double[] { 5.1, 3.5, 1.4, 0.2 };

DisplayVectorArray(inputArray, 4, 3, true);

double[] resultArray = neuralNetwork.ComputeOutputNodeValueArray(inputArray);

Console.WriteLine();

Console.WriteLine("출력 데이터 : ");

DisplayVectorArray(resultArray, 3, 3, true);

Console.WriteLine();

Console.WriteLine("신경망 훈련 데모 종료");

Console.ReadLine();

}

#endregion

#region 데이터 준비하기 - PrepareData(sourceValueArray, seed, trainingValueArray, testValueArray)

/// <summary>

/// 데이터 준비하기

/// </summary>

/// <param name="sourceValueArray">소스 값 배열</param>

/// <param name="seed">시드</param>

/// <param name="trainingValueArray">훈련 값 배열</param>

/// <param name="testValueArray">테스트 값 배열</param>

private static void PrepareData

(

double[][] sourceValueArray,

int seed,

out double[][] trainingValueArray,

out double[][] testValueArray

)

{

Random random = new Random(seed);

int rowCount = sourceValueArray.Length;

int columnCount = sourceValueArray[0].Length;

int trainingRowCount = (int)(rowCount * 0.80);

int testRowCount = rowCount - trainingRowCount;

trainingValueArray = new double[trainingRowCount][];

testValueArray = new double[testRowCount][];

double[][] copyValueArray = new double[sourceValueArray.Length][];

for(int i = 0; i < copyValueArray.Length; i++)

{

copyValueArray[i] = sourceValueArray[i];

}

for(int i = 0; i < copyValueArray.Length; i++)

{

int randomIndex = random.Next(i, copyValueArray.Length);

double[] temporaryArray = copyValueArray[randomIndex];

copyValueArray[randomIndex] = copyValueArray[i];

copyValueArray[i ] = temporaryArray;

}

for(int i = 0; i < trainingRowCount; i++)

{

trainingValueArray[i] = new double[columnCount];

for(int j = 0; j < columnCount; j++)

{

trainingValueArray[i][j] = copyValueArray[i][j];

}

for(int i = 0; i < testRowCount; i++)

{

testValueArray[i] = new double[columnCount];

for(int j = 0; j < columnCount; j++)

{

testValueArray[i][j] = copyValueArray[i + trainingRowCount][j];

}

#endregion

#region 벡터 배열 출력하기 - DisplayVectorArray(sourceVectorArray, valueCountPerRow, decimalCount, newLine)

/// <summary>

/// 벡터 배열 출력하기

/// </summary>

/// <param name="sourceVectorArray">소스 벡터 배열</param>

/// <param name="valueCountPerRow">행당 값 카운트</param>

/// <param name="decimalCount">소수점 카운트</param>

/// <param name="newLine">개행 여부</param>

private static void DisplayVectorArray(double[] sourceVectorArray, int valueCountPerRow, int decimalCount, bool newLine)

{

int sourceVectorCount = sourceVectorArray.Length;

for(int i = 0; i < sourceVectorCount; i++)

{

if(i % valueCountPerRow == 0)

{

Console.WriteLine("");

}

Console.Write(sourceVectorArray[i].ToString("F" + decimalCount).PadLeft(decimalCount + 4) + " ");

}

if (newLine == true)

{

Console.WriteLine("");

}

#endregion

#region 매트릭스 배열 출력하기 - DisplayMatrixArray(sourceMatrixArray, rowCount, decimalCount, newLine)

/// <summary>

/// 매트릭스 배열 출력하기

/// </summary>

/// <param name="sourceMatrixArray">소스 매트릭스 배열</param>

/// <param name="rowCount">행 카운트</param>

/// <param name="decimalCount">소수점 카운트</param>

/// <param name="newLine">개행 여부</param>

private static void DisplayMatrixArray(double[][] sourceMatrixArray, int rowCount, int decimalCount, bool newLine)

{

for(int i = 0; i < rowCount; i++)

{

Console.Write(i.ToString().PadLeft(3) + " : ");

for(int j = 0; j < sourceMatrixArray[i].Length; j++)

{

if(sourceMatrixArray[i][j] >= 0d)

{

Console.Write(" ");

}

else

{

Console.Write("-");

}

Console.Write(Math.Abs(sourceMatrixArray[i][j]).ToString("F" + decimalCount) + " ");

}

Console.WriteLine("");

}

if(newLine == true)

{

Console.WriteLine("");

}

#endregion

}

TestProject.zip