def getTestResult(newModel, testI, testSizeOnce):
# test all the data at once
if testSizeOnce == 0:
print('test all rows at once')
testO = deepLearning_GPU.modelOutput(newModel, testI)
testO = testO[len(testO) - 1]
# test (testSizeOnce) rows at once
else:
testSize = len(testI) # the number of entire rows
times = int((testSize - 1) / testSizeOnce + 1)
testO = [] # final output to return
# add each output of input testI_ to testO
for i in range(times):
print(str((i + 1) * testSizeOnce) + ' / ' + str(testSize))
testI_ = testI[i * testSizeOnce:min(len(testI), (i + 1) *
testSizeOnce)]
testO_ = deepLearning_GPU.modelOutput(newModel, testI_)
testO_ = testO_[len(testO_) - 1]
testO += list(testO_)
# convert to numpy array
testO = np.array(testO)
# return test result
return testO
def trainAndTest(lr, dropout):
global ALLTEXT
# training and test inputs and outputs
trainI = []
trainO = []
testI = []
testO = []
for i in range(len(inputs)):
if random.random() < 0.2:
testI.append(inputs[i])
testO.append(outputs[i])
else:
trainI.append(inputs[i])
trainO.append(outputs[i])
# model design
NN = [
tf.keras.layers.Flatten(input_shape=(len(inputs[0]), )),
keras.layers.Dense(8, activation='relu'),
keras.layers.Dropout(dropout),
keras.layers.Dense(8, activation='relu'),
keras.layers.Dropout(dropout),
keras.layers.Dense(8, activation='relu'),
keras.layers.Dense(len(outputs[0]), activation='sigmoid')
]
op = tf.keras.optimizers.Adam(lr)
# learning
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error', trainI, trainO,
'test', 50, False, True, deviceName)
# test
newModel = deepLearning_GPU.deepLearningModel('test', True)
testOutput = deepLearning_GPU.modelOutput(newModel, testI)
# estimate
outputLayer = testOutput[len(testOutput) - 1]
avgSquareError = 0.0
for i in range(len(testO)):
op = outputLayer[i] # output layer output
to = testO[i] # test output
squareError = pow(op[0] - to[0], 2) + pow(op[1] - to[1], 2)
avgSquareError += squareError
avgSquareError /= len(testO)
toAppend = str(round(lr, 8)) + ' ' + str(round(dropout, 8)) + ' ' + str(
round(avgSquareError, 8))
ALLTEXT += toAppend + '\n'
print(toAppend)
return avgSquareError
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(1, activation='sigmoid')
]
op = tf.keras.optimizers.Adam(0.0005) # learning rate = 0.0005
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error',
trainInputs, trainOutputs,
'mainCOVID19_' + str(epoch), epoch,
False, True, deviceName)
# test
print('\n <<< TEST >>>\n')
newModel = deepLearning_GPU.deepLearningModel(
'mainCOVID19_' + str(epoch), True)
testOutputs = deepLearning_GPU.modelOutput(newModel, testInputs)
# print output layer of test result
print('\n <<< test result >>>\n')
difSum = 0 # sum of difference (absolute)
valOutputSum = 0 # sum of validation output (absolute)
result = testOutputs[len(testOutputs) - 1]
binary = [
0, 0, 0, 0
] # True-positive, True-negative, False-positive, False-negative
for i in range(len(result)):
result[i] = h.invSigmoid(result[i], inf) # invSigmoid results
def deepLearning(inputFileName, outputFileName, testFileName,
testOutputFileName, imgHeight, deviceName, epoch, printed,
modelName):
# read files
trainI = helper.getDataFromFile(inputFileName,
imgHeight) # input train data
trainO = helper.getDataFromFile(
outputFileName, None) # output train data (Sigmoid applied)
testI = helper.getDataFromFile(
testFileName, imgHeight) # test input data (set nullValue to 0)
# apply sigmoid to train output data
for i in range(len(trainO)):
for j in range(len(trainO[0])):
trainO[i][j] = helper.sigmoid(trainO[i][j])
# flatten trainI: (N, size, size) -> (N, size*size)
for i in range(len(trainI)):
trainI[i] = helper.flatten(trainI[i])
print('')
print(' ---- number of rows ----')
print('input size: ' + str(len(trainI)))
print('output size: ' + str(len(trainO)))
print('test size: ' + str(len(testI)))
print('')
# print input, output, and test data
if printed != 0:
print('\n ---- original input data ----\n')
for i in range(len(trainI)):
print(helper.roundedArray(trainI[i], 6))
print('\n ---- original output data ----\n')
for i in range(len(trainO)):
print(helper.roundedArray(trainO[i], 6))
print('\n ---- original test data ----\n')
for i in range(len(testI)):
print(helper.roundedArray(testI[i], 6))
# model design using deepLearning_model.txt, in the form of
# activation function of final layer is always 'sigmoid'
f = open('deepLearning_model.txt', 'r')
modelInfo = f.readlines()
f.close()
# NN and optimizer
NN = helper.getNN(modelInfo, trainI, trainO) # Neural Network
op = helper.getOptimizer(modelInfo) # optimizer
try: # try reading test.h5 and test.json
newModel = deepLearning_GPU.deepLearningModel(modelName, True)
testOutput = deepLearning_GPU.modelOutput(newModel, testI)
except: # do learning if test.h5 and test.json does not exist
print('\n <<<< LEARNING >>>>\n')
# False, True는 각각 dataPrint(학습데이터 출력 여부), modelPrint(model의 summary 출력 여부)
print(trainO[0])
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error', trainI,
trainO, modelName, epoch, False, True,
deviceName)
newModel = deepLearning_GPU.deepLearningModel(modelName, True)
testOutput = deepLearning_GPU.modelOutput(newModel, testI)
# test
print('\n <<<< TEST >>>>\n')
# estimate
outputLayer = testOutput[len(testOutput) - 1]
# inverse sigmoid
for i in range(len(outputLayer)): # for each output data
for j in range(len(outputLayer[0])): # for each value of output data
outputLayer[i][j] = helper.invSigmoid(outputLayer[i][j])
# write to file
result = ''
print('\n<<<< output layer >>>>')
for i in range(len(outputLayer)): # for each output data
for j in range(len(outputLayer[0])): # for each value of output data
result += str(outputLayer[i][j]) + '\t'
result += '\n'
print(result)
f = open(testOutputFileName.split('.')[0] + '_prediction.txt', 'w')
f.write(result)
f.close()
def trainAndTest(numOfTrain, numOfTest, virusProb, boardSize, maxDist, deviceName): # 학습 데이터를 읽어서 배열에 저장 및 학습
# 학습 데이터 (각 x열에 해당하는 8개의 배열, y열에 해당하는 배열)
# 0 ~ 3 : hop1Noise_, hop2AvgNoise_, hop1NoiseZ_, hop2AvgNoiseZ_,
# 4 ~ 7 : hop1Max_, hop2AvgMax_, hop1MaxZ_, hop2AvgMaxZ_,
# 8 : isMal_
mainData_train = [[], [], [], [], [], [], [], [], []] # 학습 데이터
mainData_test = [[], [], [], [], [], [], [], [], []] # 테스트 데이터
# 각 데이터를 랜덤하게 학습데이터(0) 또는 테스트데이터(1) 로 지정
dataInfo = []
dataSize = numOfTrain + numOfTest # 전체 데이터의 개수
for i in range(dataSize): dataInfo.append(1)
trainCount = 0
while trainCount < numOfTrain: # 학습데이터(0) 지정
index = math.floor(random.random() * dataSize)
if dataInfo[index] != 0: # 기존에 학습데이터로 지정되지 않음
trainCount += 1
dataInfo[index] = 0
# 학습 및 테스트 데이터 파일 존재 여부와 무관하게 실행
try:
# 학습 데이터를 mainFunc을 이용하여 생성
mF = mainFunc(numOfTrain, numOfTest, virusProb, boardSize, maxDist)
mainData = mF[0] # 학습 및 테스트 데이터의 각 line
countList = mF[1] # 파일의 각 line에 대한 데이터 번호 (dataNo)
# 숫자 데이터를 float으로 변환
for i in range(len(mainData)):
for j in range(8): mainData[j][i] = float(mainData[j][i])
# numOfTrain_NB, numOfTrain_SVM, numOfTest에 추가
for i in range(len(mainData[0])):
dataNoOfThisLine = countList[i] # 데이터 파일의 해당 line의 dataNo 값
if dataInfo[dataNoOfThisLine] == 0: # 0 -> 학습 데이터
for j in range(9): mainData_train[j].append(mainData[j][i])
elif dataInfo[dataNoOfThisLine] == 1: # 1 -> 테스트 데이터
for j in range(9): mainData_test[j].append(mainData[j][i])
except Exception as e:
print('\nerror: ' + str(e) + '\n')
# 학습 데이터, 테스트 데이터의 실제 line 개수 (각 데이터마다 node 개수만큼의 line이 있음)
lenTrain = len(mainData_train[0])
lenTest = len(mainData_test[0])
print('\n################################')
print(lenTrain, lenTest, virusProb, 'boardSize:', boardSize, 'maxDist:', maxDist)
print('dataInfo: ' + str(dataInfo))
print('################################\n')
# isMal_의 'True', 'False'를 숫자 0(False), 1(True)로 변환
for i in range(lenTrain):
if mainData_train[8][i] == True: mainData_train[8][i] = 1
else: mainData_train[8][i] = 0
for i in range(lenTest):
if mainData_test[8][i] == True: mainData_test[8][i] = 1
else: mainData_test[8][i] = 0
NN = [tf.keras.layers.Flatten(input_shape=(4,)),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(1, activation='sigmoid')]
op = tf.keras.optimizers.Adam(0.001)
# 학습 데이터의 입력, 출력 데이터 생성
# 입력 데이터: hop1NoiseZ_(2), hop2AvgNoiseZ_(3), hop1MaxZ_(6), hop2AvgMaxZ_(7) / 출력 데이터: isMal_(8)
inputs = []
outputs = []
for i in range(lenTrain):
inputs.append([mainData_train[2][i], mainData_train[3][i], mainData_train[6][i], mainData_train[7][i]])
outputs.append([mainData_train[8][i]])
# 테스트 데이터의 입력 데이터 생성
testInputs = [] # 테스트 입력
testValOutputs = [] # 테스트에 대한 validation data (실제 결과값)
for i in range(lenTest):
testInputs.append([mainData_test[2][i], mainData_test[3][i], mainData_test[6][i], mainData_test[7][i]])
testValOutputs.append(mainData_test[8][i])
# 학습 실시
print(inputs[0:5])
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error', inputs, outputs, '200407_DTN', 2500, False, True, deviceName)
# 테스트 실시
testModel = deepLearning_GPU.deepLearningModel('200407_DTN', True)
testOutputAllLayers = deepLearning_GPU.modelOutput(testModel, testInputs)
testOutput = testOutputAllLayers[len(testOutputAllLayers)-1] # 모델에 testInputs를 입력했을 때의 결과값
# 오차 제곱의 평균 계산
difAvg = 0
for i in range(lenTest): difAvg += pow(testOutput[i][0] - testValOutputs[i], 2)
difAvg /= lenTest
print('mean square error: ' + str(difAvg))
print('')
# TP, TN, FP, FN 계산
TP = 0 # True Positive
TN = 0 # True Negative
FP = 0 # False Positive
FN = 0 # False Negative
for i in range(lenTest):
if testOutput[i][0] >= 0.5 and testValOutputs[i] >= 0.5: TP += 1
elif testOutput[i][0] < 0.5 and testValOutputs[i] < 0.5: TN += 1
elif testOutput[i][0] >= 0.5 and testValOutputs[i] < 0.5: FP += 1
elif testOutput[i][0] < 0.5 and testValOutputs[i] >= 0.5: FN += 1
ALL = TP + TN + FP + FN
RealP = TP + FN
RealN = TN + FP
correct = TP + FP
# 테스트 결과 출력
print('True Positive : ' + str(TP) + ' / ' + str(round(100 * TP / ALL, 2)) + '%')
print('True Negative : ' + str(TN) + ' / ' + str(round(100 * TN / ALL, 2)) + '%')
print('False Positive : ' + str(FP) + ' / ' + str(round(100 * FP / ALL, 2)) + '%')
print('False Negative : ' + str(FN) + ' / ' + str(round(100 * FN / ALL, 2)) + '%')
print('')
print('Real Positive : ' + str(RealP) + ' / ' + str(round(100 * RealP / ALL, 2)) + '%')
print('Real Negative : ' + str(RealN) + ' / ' + str(round(100 * RealN / ALL, 2)) + '%')
print('')
print('correct : ' + str(correct) + ' / ' + str(round(100 * correct / ALL, 2)) + '%')
print('')
def trainAndTest(numOfTrain, numOfTest, virusProb, boardSize, maxDist,
deviceName): # 학습 데이터를 읽어서 배열에 저장 및 학습
# 학습 및 테스트 데이터 파일 존재 여부와 무관하게 실행
# 학습 데이터를 mainFunc을 이용하여 생성
mF = mainFunc(numOfTrain, numOfTest, virusProb, boardSize, maxDist)
mainData = mF[0] # 학습 및 테스트 데이터의 각 line
countList = mF[1] # 파일의 각 line에 대한 데이터 번호 (dataNo)
# 학습 실시
#print(inputs[0:5])
epochList = input('epochs:')
testResult = ''
testTable = 'epoch\ttrain\ttest\terror\tTP\tTPrate\tTN\tTNrate\tFP\tFPrate\tFN\tFNrate\treal P\trP rate\treal N\trN rate\tcorrect\tco rate\n'
for epoch in range(len(epochList.split(' '))):
# 학습 데이터 (각 x열에 해당하는 8개의 배열, y열에 해당하는 배열)
# 0 ~ 5 : hop1Noise_, hop2AvgNoise_, hop2AvgNoise_, hop2AvgMax_, hop1Neis_, hop2AvgNeis_
# 6 ~11 : hop1NoiseZ_, hop1MaxZ_, hop2AvgNoiseZ_, hop2AvgMaxZ_, hop1NeisZ_, hop2AvgNeisZ_,
# 12 : isMal_
mainData_train = [[], [], [], [], [], [], [], [], [], [], [], [],
[]] # 학습 데이터
mainData_test = [[], [], [], [], [], [], [], [], [], [], [], [],
[]] # 테스트 데이터
# 각 데이터를 랜덤하게 학습데이터(0) 또는 테스트데이터(1) 로 지정
dataInfo = []
dataSize = numOfTrain + numOfTest # 전체 데이터의 개수
for i in range(dataSize):
dataInfo.append(1)
trainCount = 0
while trainCount < numOfTrain: # 학습데이터(0) 지정
index = math.floor(random.random() * dataSize)
if dataInfo[index] != 0: # 기존에 학습데이터로 지정되지 않음
trainCount += 1
dataInfo[index] = 0
# 숫자 데이터를 float으로 변환
for i in range(len(mainData)):
for j in range(12):
mainData[j][i] = float(mainData[j][i])
# numOfTrain_NB, numOfTrain_SVM, numOfTest에 추가
for i in range(len(mainData[0])):
dataNoOfThisLine = countList[i] # 데이터 파일의 해당 line의 dataNo 값
if dataInfo[dataNoOfThisLine] == 0: # 0 -> 학습 데이터
for j in range(13):
mainData_train[j].append(mainData[j][i])
elif dataInfo[dataNoOfThisLine] == 1: # 1 -> 테스트 데이터
for j in range(13):
mainData_test[j].append(mainData[j][i])
# 학습 데이터, 테스트 데이터의 실제 line 개수 (각 데이터마다 node 개수만큼의 line이 있음)
lenTrain = len(mainData_train[0])
lenTest = len(mainData_test[0])
print('length:', lenTrain, lenTest)
print('\n################################')
print(lenTrain, lenTest, virusProb, 'boardSize:', boardSize,
'maxDist:', maxDist)
print('dataInfo: ' + str(dataInfo))
print('################################\n')
# isMal_의 'True', 'False'를 숫자 0(False), 1(True)로 변환
for i in range(lenTrain):
if mainData_train[12][i] == True: mainData_train[8][i] = 1
else: mainData_train[12][i] = 0
for i in range(lenTest):
if mainData_test[12][i] == True: mainData_test[8][i] = 1
else: mainData_test[12][i] = 0
# 신경망 생성
NN = [
tf.keras.layers.Flatten(input_shape=(4, )),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(1, activation='sigmoid')
]
op = tf.keras.optimizers.Adam(0.001)
# 학습 데이터의 입력, 출력 데이터 생성
# 입력 데이터: hop1NoiseZ_, hop1MaxZ_, hop2AvgNoiseZ_, hop2AvgMaxZ_(6~9) / 출력 데이터: isMal_(12)
inputs = []
outputs = []
for i in range(lenTrain):
inputs.append([
mainData_train[6][i], mainData_train[7][i],
mainData_train[8][i], mainData_train[9][i]
])
outputs.append([mainData_train[12][i]])
# 테스트 데이터의 입력 데이터 생성
testInputs = [] # 테스트 입력
testValOutputs = [] # 테스트에 대한 validation data (실제 결과값)
for i in range(lenTest):
testInputs.append([
mainData_test[6][i], mainData_test[7][i], mainData_test[8][i],
mainData_test[9][i]
])
testValOutputs.append(mainData_test[12][i])
# 학습 실시
ep = int(epochList.split(' ')[epoch])
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error', inputs,
outputs, '200407_DTN', ep, False, True,
deviceName)
# 테스트 실시
testModel = deepLearning_GPU.deepLearningModel('200407_DTN', True)
testOutputAllLayers = deepLearning_GPU.modelOutput(
testModel, testInputs)
testOutput = testOutputAllLayers[len(testOutputAllLayers) -
1] # 모델에 testInputs를 입력했을 때의 결과값
# 오차 제곱의 평균 계산
difAvg = 0
for i in range(lenTest):
difAvg += pow(testOutput[i][0] - testValOutputs[i], 2)
difAvg /= lenTest
print('mean square error: ' + str(difAvg))
print('')
# TP, TN, FP, FN 계산
TP = 0 # True Positive
TN = 0 # True Negative
FP = 0 # False Positive
FN = 0 # False Negative
for i in range(lenTest):
if testOutput[i][0] >= 0.5 and testValOutputs[i] >= 0.5: TP += 1
elif testOutput[i][0] < 0.5 and testValOutputs[i] < 0.5: TN += 1
elif testOutput[i][0] >= 0.5 and testValOutputs[i] < 0.5: FP += 1
elif testOutput[i][0] < 0.5 and testValOutputs[i] >= 0.5: FN += 1
ALL = TP + TN + FP + FN
RealP = TP + FN
RealN = TN + FP
correct = TP + TN
# 테스트 결과 저장
testResult += ' <<< epoch ' + str(ep) + ' >>>\n'
testResult += 'mean square error: ' + str(difAvg) + ' (train: ' + str(
len(inputs)) + ' test:' + str(ALL) + ')\n\n'
testResult += 'True Positive : ' + str(TP) + ' / ' + str(
round(100 * TP / ALL, 2)) + '%\n'
testResult += 'True Negative : ' + str(TN) + ' / ' + str(
round(100 * TN / ALL, 2)) + '%\n'
testResult += 'False Positive : ' + str(FP) + ' / ' + str(
round(100 * FP / ALL, 2)) + '%\n'
testResult += 'False Negative : ' + str(FN) + ' / ' + str(
round(100 * FN / ALL, 2)) + '%\n\n'
testResult += 'Real Positive : ' + str(RealP) + ' / ' + str(
round(100 * RealP / ALL, 2)) + '%\n'
testResult += 'Real Negative : ' + str(RealN) + ' / ' + str(
round(100 * RealN / ALL, 2)) + '%\n'
testResult += 'correct : ' + str(correct) + ' / ' + str(
round(100 * correct / ALL, 2)) + '%\n\n'
testTable += str(ep) + '\t' + str(
len(inputs)) + '\t' + str(ALL) + '\t' + str(round(
difAvg, 4)) + '\t' + str(TP) + '\t' + str(
round(100 * TP / ALL, 2)) + '%\t'
testTable += str(TN) + '\t' + str(round(100 * TN / ALL, 2)) + '%\t'
testTable += str(FP) + '\t' + str(round(100 * FP / ALL, 2)) + '%\t'
testTable += str(FN) + '\t' + str(round(100 * FN / ALL, 2)) + '%\t'
testTable += str(RealP) + '\t' + str(round(100 * RealP / ALL,
2)) + '%\t'
testTable += str(RealN) + '\t' + str(round(100 * RealN / ALL,
2)) + '%\t'
testTable += str(correct) + '\t' + str(round(100 * correct / ALL,
2)) + '%\n'
# 파일에 쓰기
tR = open('DTN_testResult.txt', 'w')
tR.write(testResult)
tR.close()
tT = open('DTN_testTable.txt', 'w')
tT.write(testTable)
tT.close()
NN = [
tf.keras.layers.Flatten(input_shape=(2, )),
keras.layers.Dense(16, activation='relu'),
keras.layers.Dense(16, activation='relu'),
keras.layers.Dense(2, activation='sigmoid')
]
op = tf.keras.optimizers.Adam(0.001)
inputs = [[2, 1], [3, 2], [5, 3], [8, 5], [13, 8], [21, 13], [34, 21]]
outputs = [[0.2, 0.5], [0.4, 0.6], [0.5, 0.7], [0.7, 0.75], [0.8, 0.77],
[0.85, 0.8], [0.9, 0.85]]
# learning
print('\n ' + ('#===' * 16) + ' <<<< LEARNING >>>> ' + ('===#' * 16) +
'\n')
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error', inputs,
outputs, 'test', 500, True, True, deviceName)
# test
print('\n ' + ('#===' * 16) + ' <<<< TEST >>>> ' + ('===#' * 16) + '\n')
print('\n << test output (첫번째 Neural Network) >>\n')
newModel = deepLearning_GPU.deepLearningModel('test', True)
testOutput = deepLearning_GPU.modelOutput(newModel,
[[4, 2.5], [6, 3.5], [7, 4.5]])
print('\n[[4, 2.5], [6, 3.5], [7, 4.5]]에 대한 학습 결과:\n')
for i in range(len(testOutput)):
print(' << layer No: ' + str(i) + ' >>')
print(str(testOutput[i]))
def deepLearning(inputFileName, outputFileName, testFileName,
testOutputFileName, valid, deviceName, epoch, printed,
modelName, normalizeTarget):
# read files
# trainO : (originalOutput - meanOriginalOutput)/stdOriginalOutput
trainI = helper.getDataFromFile(inputFileName, None) # input train data
trainO = helper.getDataFromFile(
outputFileName, None) # output train data (Sigmoid applied)
testI = helper.getDataFromFile(
testFileName, None) # test input data (set nullValue to 0)
# apply sigmoid to train output data
# trainO : sigmoid(normalize(originalOutput))
# = sigmoid((originalOutput - meanOriginalOutput)/stdOriginalOutput)
for i in range(len(trainO)):
for j in range(len(trainO[0])):
trainO[i][j] = helper.sigmoid(trainO[i][j])
# for i in range(15): print(trainO[i])
print('')
print(' ---- number of rows ----')
print('input size: ' + str(len(trainI)))
print('output size: ' + str(len(trainO)))
print('test size: ' + str(len(testI)))
print('')
# print input, output, and test data
if printed != 0:
print('\n ---- original input data ----\n')
for i in range(len(trainI)):
print(helper.roundedArray(trainI[i], 6))
print('\n ---- original output data ----\n')
for i in range(len(trainO)):
print(helper.roundedArray(trainO[i], 6))
print('\n ---- original test data ----\n')
for i in range(len(testI)):
print(helper.roundedArray(testI[i], 6))
# model design using deepLearning_model.txt, in the form of
# activation function of final layer is always 'sigmoid'
f = open('deepLearning_model.txt', 'r')
modelInfo = f.readlines()
f.close()
# read normalization info
if normalizeTarget == True:
fnorm = open('data_normalizeInfo.txt', 'r')
fnormInfo = fnorm.readlines()
fnormMean = float(fnormInfo[0].split(' ')[0]) # mean of training data
fnormStd = float(fnormInfo[0].split(' ')[1]) # stddev of training data
#### TEST when the value of valid is 0 ####
if valid == 0:
# NN and optimizer
NN = helper.getNN(modelInfo, trainI, trainO) # Neural Network
op = helper.getOptimizer(modelInfo) # optimizer
#print(trainI[:5])
#print(trainO[:5])
try: # try reading test.h5 and test.json
newModel = deepLearning_GPU.deepLearningModel(modelName, True)
testOutput = deepLearning_GPU.modelOutput(newModel, testI)
except: # do learning if test.h5 and test.json does not exist
print('\n <<<< LEARNING >>>>\n')
# False, True는 각각 dataPrint(학습데이터 출력 여부), modelPrint(model의 summary 출력 여부)
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error', trainI,
trainO, modelName, epoch, False,
True, deviceName)
newModel = deepLearning_GPU.deepLearningModel(modelName, True)
testOutput = deepLearning_GPU.modelOutput(newModel, testI)
# test
print('\n <<<< TEST >>>>\n')
# estimate
outputLayer = testOutput[len(testOutput) - 1]
# inverse sigmoid
# output: denormalize(invSigmoid(sigmoid(normalize(originalOutput))))
# = denormalize((originalOutput - meanOriginalOutput)/stdOriginalOutput)
# = originalOutput
for i in range(len(outputLayer)): # for each output data
for j in range(len(
outputLayer[0])): # for each value of output data
outputLayer[i][j] = helper.invSigmoid(outputLayer[i][j])
if normalizeTarget == True:
outputLayer[i][
j] = outputLayer[i][j] * fnormStd + fnormMean
# write to file
result = ''
print('\n<<<< output layer >>>>')
for i in range(len(outputLayer)): # for each output data
for j in range(len(
outputLayer[0])): # for each value of output data
result += str(outputLayer[i][j]) + '\t'
result += '\n'
f = open(testOutputFileName.split('.')[0] + '_prediction.txt', 'w')
f.write(result)
f.close()
# return final result
finalResult = []
for i in range(len(outputLayer)): # for each output data
finalResult.append(outputLayer[i][0])
return finalResult
#### VALIDATION when the value of valid is >0 ####
else:
# make index-list of validation data
inputSize = len(trainI)
validSize = int(inputSize * valid)
trainSize = inputSize - validSize
validArray = []
for i in range(inputSize):
validArray.append(0)
while sum(validArray) < validSize:
validArray[random.randint(0, inputSize - 1)] = 1
# make train and validation data
# _TrainO, _ValidO : sigmoid((originalOutput - meanOriginalOutput)/stdOriginalOutput)
_TrainI = [] # training input
_TrainO = [] # training output
_ValidI = [] # valid input
_ValidO = [] # valid output
for i in range(inputSize):
if validArray[i] == 0: # training data
_TrainI.append(trainI[i])
_TrainO.append(trainO[i])
else: # validation data
_ValidI.append(trainI[i])
_ValidO.append(trainO[i])
# model name for validation
newModelName = modelName + 'Valid'
# NN and optimizer
NN = helper.getNN(modelInfo, _TrainI, _TrainO) # Neural Network
op = helper.getOptimizer(modelInfo) # optimizer
# output for validation
try: # try reading testValid.h5 and test.json
validModel = deepLearning_GPU.deepLearningModel(newModelName, True)
predictedValidO = deepLearning_GPU.modelOutput(validModel, _ValidI)
except: # do learning if testValid.h5 and test.json does not exist
print('\n <<<< LEARNING >>>>\n')
# False, True는 각각 dataPrint(학습데이터 출력 여부), modelPrint(model의 summary 출력 여부)
# _TrainO : sigmoid((originalOutput - meanOriginalOutput)/stdOriginalOutput)
deepLearning_GPU.deepLearning(NN, op, 'mean_squared_error',
_TrainI, _TrainO, newModelName,
epoch, False, True, deviceName)
validModel = deepLearning_GPU.deepLearningModel(newModelName, True)
predictedValidO = deepLearning_GPU.modelOutput(validModel, _ValidI)
# evaluation
print('\n <<<< VALID >>>>\n')
MAE = 0 # mean absolute error
MSE = 0 # mean square error
accuracy = 0 # accuracy
# predicted validation output
outputLayer = predictedValidO[len(predictedValidO) - 1]
# inverse sigmoid
# output : invSigmoid(sigmoid(normalize(originalOutput)))
# = (originalOutput - meanOriginalOutput)/stdOriginalOutput
for i in range(len(outputLayer)): # for each output data
for j in range(len(
outputLayer[0])): # for each value of output data
outputLayer[i][j] = helper.invSigmoid(outputLayer[i][j])
# compute error
# output : denormalize((originalOutput - meanOriginalOutput)/stdOriginalOutput)
# = originalOutput
# _Valid0 : denormalize(invSigmoid(sigmoid((originalOutput - meanOriginalOutput)/stdOriginalOutput)))
# = denormalize((originalOutput - meanOriginalOutput)/stdOriginalOutput)
# = originalOutput
for i in range(len(outputLayer)): # for each output data
for j in range(len(
outputLayer[0])): # for each value of output data
_ValidO[i][j] = helper.invSigmoid(_ValidO[i][j])
if normalizeTarget == True:
_ValidO[i][j] = _ValidO[i][j] * fnormStd + fnormMean
outputLayer[i][
j] = outputLayer[i][j] * fnormStd + fnormMean
# compute error
validCount = 0
resultToWrite = ''
for i in range(inputSize):
if validArray[i] == 1:
# compute errors and accuracy
thisAE = abs(_ValidO[validCount][0] -
outputLayer[validCount][0])
thisSE = pow(
_ValidO[validCount][0] - outputLayer[validCount][0], 2)
MAE += thisAE
MSE += thisSE
if thisSE <= 0.5: accuracy += 1
# print and write result
newResultToWrite = ('[' + str(i) + '] pred = ' +
str(int(outputLayer[validCount][0])) +
', real = ' +
str(int(_ValidO[validCount][0])) +
', AE = ' + str(int(thisAE)) + ', SE = ' +
str(int(thisSE)))
resultToWrite += newResultToWrite + '\n'
print(newResultToWrite)
validCount += 1
MAE /= validSize
MSE /= validSize
accuracy /= validSize
# print evaluation result
resultSummary = ''
resultSummary += 'input size : ' + str(inputSize) + '\n'
resultSummary += 'train size : ' + str(trainSize) + '\n'
resultSummary += 'valid size : ' + str(validSize) + '\n'
resultSummary += 'MAE : ' + str(round(MAE, 6)) + '\n'
resultSummary += 'MSE : ' + str(round(MSE, 6)) + '\n'
resultSummary += 'accuracy : ' + str(round(accuracy, 6)) + '\n'
resultSummary += 'pred avg : ' + str(np.average(outputLayer,
axis=0)) + '\n'
resultSummary += 'real avg : ' + str(np.average(_ValidO,
axis=0)) + '\n'
print(resultSummary)
resultToWrite += resultSummary
# write result file
fvalid = open('data_valid_result.txt', 'w')
fvalid.write(resultToWrite)
fvalid.close()
def run(size, numTrain, numTest, numWD, deviceName, doTrainAndTest):
inputs = [] # 딥러닝 입력값 저장
outputs = [] # 딥러닝 출력값 저장
isFile = False # valueMaze_X.json 학습 모델 파일이 있는가?
# 입력 정보
originalScreen = [] # 각 맵의 스크린
wdList = [
] # 각 맵에서 wireless device의 위치를 저장한 배열, [[wd0Y, wd0X], [wd1Y, wd1X], ..., [wd(n-1)Y, wd(n-1)X]]
# 출력 정보
outputScreen = [
] # 각 맵에서의 각 좌표에서의 throughput을 나타낸 맵, [[[thrput(0,0), thrput(0,1), ...], [thrput(1,0), ...], ...], (so on)]
# 평가 결과 저장
RL = []
# 1. 트레이닝 및 테스트용 맵 생성
# 맵의 구성: .(공간), H(HAP), W(wireless device)
if readOrWrite == 1: # 맵 파일 쓰고 그 맵으로 트레이닝, 테스트
print('generating maps...')
for i in range(numTrain + numTest):
initScreen_ = WPCN_helper_REAL.initScreen(size, False, numWD)
originalScreen.append(initScreen_[0])
wdList.append(initScreen_[2])
# 맵 파일 작성
f = open(
'originalMaps_' + str(size) + '_' + str(numWD) + '/DL_WPCN_' +
('0' if i < 1000 else '') + ('0' if i < 100 else '') +
('0' if i < 10 else '') + str(i) + '.txt', 'w')
for j in range(size):
for k in range(size):
if originalScreen[i][j][k] == -1:
f.write('W') # wireless device
elif originalScreen[i][j][k] == 0:
f.write('.') # space
elif originalScreen[i][j][k] == 1:
f.write('H') # HAP (not exist in the original map)
f.write('\n')
f.close()
else: # 기존 맵 파일 읽어서 기존 맵으로 트레이닝, 테스트
print('reading maps...')
for i in range(numTrain + numTest):
f = open(
'originalMaps_' + str(size) + '_' + str(numWD) + '/DL_WPCN_' +
('0' if i < 1000 else '') + ('0' if i < 100 else '') +
('0' if i < 10 else '') + str(i) + '.txt', 'r')
map_ = f.readlines() # 맵을 나타낸 배열
f.close()
for j in range(len(map_)):
map_[j] = map_[j].replace('\n', '') # 각 줄마다 개행문자 제거
# originalScreen과 wdList 읽어오기
wdListTemp = []
thisScreen = [[0] * size
for j in range(size)] # originalScreen에 추가할 맵(스크린)
for j in range(size):
for k in range(size):
if map_[j][k] == 'W':
thisScreen[j][k] = -1 # wireless device
wdListTemp.append([j, k]) # wdListTemp에 추가
elif map_[j][k] == '.':
thisScreen[j][k] = 0 # space
elif map_[j][k] == 'H':
thisScreen[j][
k] = 1 # HAP (not exist in the original map)
originalScreen.append(thisScreen)
wdList.append(wdListTemp)
# print(wdList)
# 2. 트레이닝 및 테스트용 맵에서 각 x, y (HAP의 좌표)에 대해 최적의 할당 시간(HAPtime)을 찾아서,
# HAP의 각 좌표에서의 throughput에 대한 맵 만들기 (optiInfoForMap_X.txt where X = 0 or 1)
# y좌표, x좌표는 1 간격
# 형식:
# * (맵 번호) (HAPtime) (최적y) (최적x) (최대thrput)
# (thrput at y=0,x=0) (thrput at y=0,x=1) ... (thrput at y=0,x=n)
# (thrput at y=1,x=0) (thrput at y=1,x=1) ... (thrput at y=1,x=n)
# ... ... ...
# (thrput at y=n,x=0) (thrput at y=n,x=1) ... (thrput at y=n,x=n)
toSave = ''
lines = 0 # optiInfoForMap_X.txt 파일의 라인 개수 (X = 0 or 1)
try: # optiInfoForMap 파일이 있으면 읽기
f = open(
'optiInfoForMap/optiInfoForMap_' + str(problemNo) + '_forPaper_' +
str(size) + '_' + str(numWD) + '.txt', 'r')
optiInformation = f.readlines()
f.close()
lines = len(optiInformation) # 데이터 개수는 lines / (1 + size)
if lines / (1 + size) < numTrain + numTest: # 완전한 정보가 없으면
for i in range(lines):
toSave += optiInformation[i] # toSave에 기존 저장된 정보 추가
raiseError = 1 / 0 # 오류 발생시키기
# 출력값 배열에 기존 파일에 있는 추가
temp = [] # 출력값 배열에 추가할 데이터
for i in range(lines):
# '*' 기호 (새로운 맵)를 만나면 기존 temp 배열의 데이터를 저장하고 temp는 초기화
if optiInformation[i][0] == '*':
if len(temp) > 0: outputScreen.append(temp)
temp = []
# 그렇지 않으면 temp에 해당 행의 데이터를 추가
elif len(optiInformation[i]) >= 3: # 공백 또는 개행만 있는 줄 제외
temp.append(optiInformation[i].split(' '))
for j in range(len(temp[len(temp) - 1])):
temp[len(temp) - 1][j] = float(temp[len(temp) - 1][j])
except Exception as e: # optiInfoForMap 파일이 없으면 새로 생성하기
print(e)
print("can't read optiInfoForMap_" + str(problemNo) + '_forPaper_' +
str(size) + '_' + str(numWD) + '.txt')
for i in range(int(lines / (1 + size)), numTrain + numTest):
print('finding max throughput for map ' + str(i) + '...')
# HAP의 위치(각 좌표)에 따른 최적의 할당 시간을 찾아서 출력값에 추가
temp = [
[0] * size for _ in range(size)
] # 출력값 배열에 추가할 데이터 (해당 x, y 좌표에서의 HAPtime에 따른 최대 throughput)
optiY_ = 0 # throughput (sum 또는 common)이 최대가 되는 HAP의 y좌표
optiX_ = 0 # throughput (sum 또는 common)이 최대가 되는 HAP의 x좌표
maxThroughput_ = 0.0 # throughput (sum 또는 common)의 최댓값
maxHAPtime_ = 0.0 # throughput (sum 또는 common)이 최대가 되기 위한 HAP에 할당되는 시간
for y in range(size):
for x in range(size):
(throughput, HAPtime) = WPCN_helper_REAL.getThroughput(
wdList[i], [y, x], size, problemNo)
temp[y][x] = throughput
# throughput 최고기록을 갱신한 경우 업데이트
if throughput > maxThroughput_:
optiY_ = y
optiX_ = x
maxThroughput_ = throughput
maxHAPtime_ = HAPtime
# toSave에 추가
# * (맵 번호) (HAPtime) (최적y) (최적x) (최대thrput)
# (thrput at y=0,x=0) (thrput at y=0,x=1) ... (thrput at y=0,x=n)
# (thrput at y=1,x=0) (thrput at y=1,x=1) ... (thrput at y=1,x=n)
# ... ... ...
# (thrput at y=n,x=0) (thrput at y=n,x=1) ... (thrput at y=n,x=n)
toSave += '* ' + str(i) + ' ' + str(maxHAPtime_) + ' ' + str(
optiY_) + ' ' + str(optiX_) + ' ' + str(maxThroughput_) + '\n'
for j in range(size):
for k in range(size - 1):
toSave += str(temp[j][k]) + ' '
toSave += str(temp[j][size - 1]) + '\n'
# 출력 및 toSave에 추가
print('max throughput: ' + str(maxThroughput_) + ' axis: ' +
str(optiY_) + ' ' + str(optiX_) + ' HAPtime: ' +
str(maxHAPtime_))
# 출력값 배열에 추가
if i < numTrain: outputScreen.append(temp)
# 파일로 저장
optiInfo = open(
'optiInfoForMap/optiInfoForMap_' + str(problemNo) +
'_forPaper_' + str(size) + '_' + str(numWD) + '.txt', 'w')
optiInfo.write(toSave)
optiInfo.close()
# 3. 트레이닝용 맵의 정보를 입력값과 출력값 배열에 넣기
print('make input and output data...')
for i in range(numTrain):
originalScreen_ = deepLearning_GPU_helper.arrayCopy(
originalScreen[i]) # copy array from originalScreen
originalScreen_ = deepLearning_GPU_helper.arrayCopyFlatten(
originalScreen_, 0, size, 0, size, None) # flatten
inputs.append(originalScreen_)
# find max value in outputScreen[i]
maxVal = 0.0
for j in range(size):
for k in range(size):
if outputScreen[i][j][k] > maxVal:
maxVal = outputScreen[i][j][k]
for j in range(size):
for k in range(size):
outputScreen[i][j][
k] /= maxVal # uniform [0, maxVal] -> uniform [0, 1]
outputScreen[i][j][k] = outputScreen[i][j][
k] * 2.0 - 1.0 # uniform [0, 1] -> uniform [-1, 1]
# 테스트 시 역sigmoid를 적용할 것이므로 먼저 outputScreen의 모든 원소에 sigmoid를 적용 (outputScreen의 값이 1 이상이 될수 있으므로)
for j in range(size):
for k in range(size):
outputScreen[i][j][k] = deepLearning_GPU_helper.sigmoid(
outputScreen[i][j][k])
outputs.append([outputScreen[i]])
# 트레이닝, 테스트를 하지 않고 종료
if doTrainAndTest == False: return
# 4~7. 학습 및 테스트는 주어진 입출력 데이터를 이용하여 계속 반복
while True:
epoch = input('epoch').split(
' '
) # 입력형식: epoch0 epoch1 epoch2 ... epochN (ex: 5 15 30 100 200 300 500 1000)
dropout = input('dropout').split(
' '
) # 입력형식: dropout0, dropout1, dropout2, ..., dropoutN (ex: 0 0.05 0.1 0.15 0.2)
# 딥러닝 학습시키기
lc = 'mean_squared_error' # loss calculator
for epo in range(len(epoch)): # 각 epoch에 대하여
for dro in range(len(dropout)): # 각 dropout에 대하여 실험
epoc = int(epoch[epo])
drop = float(dropout[dro])
while True:
# 실험 결과
resultForThisMap = ''
WPCNdeepNN_modelName = 'WPCNdeepNN_' + str(
size) + '_' + str(numWD)
try:
# 4. 모델 정의
NN = [
tf.keras.layers.Reshape(
(size, size, 1), input_shape=(size * size, )),
keras.layers.Conv2D(32,
kernel_size=(3, 3),
input_shape=(size, size, 1),
activation='relu'),
keras.layers.MaxPooling2D(pool_size=2),
keras.layers.Dropout(drop),
keras.layers.Conv2D(32, (3, 3), activation='relu'),
keras.layers.Flatten(),
keras.layers.Dropout(drop),
keras.layers.Dense(40, activation='relu'),
keras.layers.Dense(size * size,
activation='sigmoid'),
keras.layers.Reshape((1, size, size),
input_shape=(size * size, ))
]
# 5. 옵티마이저
if problemNo == 0:
op = tf.keras.optimizers.Adam(0.001) # for Sum-
else:
op = tf.keras.optimizers.Adam(
0.0001) # for Common-
print('[00] training...')
deepLearning_GPU.deepLearning(NN, op, lc, inputs,
outputs,
WPCNdeepNN_modelName,
epoc, False, True,
deviceName)
# 6. 테스트 데이터에 대한 출력 결과 확인하고 정답과 비교하기
print('[10] loading model...')
newModel = deepLearning_GPU.deepLearningModel(
WPCNdeepNN_modelName, True)
print('[20] init throughput values...')
sumTestThroughput = 0.0 # test throughput의 합계
sumCorrectMaxThroughput = 0.0 # 정답의 throughput의 합계 (training data를 생성할 때와 같은 방법으로 최대 throughput 확인)
optiInfo = open(
'optiInfoForMap/optiInfoForMap_' + str(problemNo) +
'_forPaper_' + str(size) + '_' + str(numWD) +
'.txt', 'r')
optiInformation = optiInfo.readlines()
optiInfo.close()
print('[30] testing...')
for i in range(numTest):
if i == 0: print('[31] initializing...')
testScreen = originalScreen[numTrain +
i] # 테스트할 스크린
testOutput = deepLearning_GPU.modelOutput(
newModel,
[np.array(testScreen).reshape(size * size)
]) # 테스트 결과
testOutputLayer = testOutput[
len(testOutput) - 1] # 테스트 결과의 output layer의 값
# 6-0. 테스트 결과 확인
# 출력값 받아오기
if i == 0: print('[32] getting optiScreen...')
optiScreen = [[0] * size for j in range(size)
] # 테스트 결과의 출력 스크린(=맵)
for j in range(size):
for k in range(size):
optiScreen[j][
k] = deepLearning_GPU_helper.invSigmoid(
testOutputLayer[0][0][j][k])
if i == 0:
print('[33] finding optimal X and Y values...')
# 출력값에서 최적의 HAP의 x, y좌표 찾기
optiY_test = 0 # 출력 map에서의 최적의 HAP의 y좌표
optiX_test = 0 # 출력 map에서의 최적의 HAP의 x좌표
optiThroughput = 0 # 출력 map에서의 최적의 HAP의 x, y좌표에서의 throughput
for j in range(size):
for k in range(size):
if optiScreen[j][k] > optiThroughput:
optiY_test = j
optiX_test = k
optiThroughput = optiScreen[j][k]
# 상하/좌우의 throughput의 값을 이용하여 좌표 보정
if i == 0:
print('[34] modifying X and Y values...')
optiY = optiY_test
optiX = optiX_test
# y좌표를 (y-1), y, (y+1) 좌표의 throughput 값을 이용하여 보정
if optiY > 0 and optiY < size - 1:
up = optiScreen[optiY - 1][optiX]
this = optiScreen[optiY][optiX]
down = optiScreen[optiY + 1][optiX]
minVal = min(up, this, down)
up -= minVal
this -= minVal
down -= minVal
optiY_test += (up * (-1) +
down * 1) / (up + this + down)
# x좌표를 (x-1), x, (x+1) 좌표의 throughput 값을 이용하여 보정
if optiX > 0 and optiX < size - 1:
left = optiScreen[optiY][optiX - 1]
this = optiScreen[optiY][optiX]
right = optiScreen[optiY][optiX + 1]
minVal = min(left, this, right)
left -= minVal
this -= minVal
right -= minVal
optiX_test += (left * (-1) + right * 1) / (
left + this + right)
# 출력값 중 HAP의 좌표를 범위 내로 조정
if i == 0:
print(
'[35] setting X and Y values as inside the range...'
)
if optiY_test > size - 1: optiY_test = size - 1
elif optiY_test < 0: optiY_test = 0
if optiX_test > size - 1: optiX_test = size - 1
elif optiX_test < 0: optiX_test = 0
# 테스트 결과 받아오기
if i == 0: print('[36] getting test results...')
(throughput,
HAPtime) = WPCN_helper_REAL.getThroughput(
wdList[numTrain + i],
[optiY_test, optiX_test], size, problemNo)
sumTestThroughput += throughput
# 6-1. testScreen과 optiScreen 중 일부 출력
if i == 0: print('[37] printing...')
if i < 10: # 처음 10개의 map에 대해서만 출력
# testScreen 출력
print(' --- test input screen for map ' +
str(i) + ' ---')
toPrint = ''
for j in range(size):
toPrint += '['
for k in range(size):
if testScreen[j][k] == -1:
toPrint += 'W '
else:
toPrint += '. '
toPrint += ']\n'
print(toPrint)
# optiScreen 출력
print(' --- test output screen for map ' +
str(i) + ' ---')
toPrint = ''
for j in range(size):
toPrint += '['
for k in range(size):
toPrint += str(
round(optiScreen[j][k], 3)
) + ' ' * (7 - len(
str(round(optiScreen[j][k], 3))))
toPrint += ']\n'
print(toPrint)
# 6-2. 정답과 비교
# 최적의 HAP의 위치 및 할당 시간 찾기
# HAP의 위치 및 할당 시간에 따른 throughput의 최댓값
if i == 0:
print('[38] comparing with correct answer...')
correctMaxThroughput = float(
optiInformation[(numTrain + i) *
(size + 1)].split(' ')[5])
sumCorrectMaxThroughput += correctMaxThroughput
# 정답과 비교
resultForThisMap += (
'test throughput for map ' + str(i) + ' [Y=' +
str(round(optiY_test, 3)) + ' X=' +
str(round(optiX_test, 3)) + ' HT=' +
str(HAPtime) + '] : ' +
str(round(throughput, 6)) + ' / ' +
str(round(correctMaxThroughput, 6)) + ' (' +
str(
round(
100.0 * throughput /
correctMaxThroughput, 6)) + ' %)\n')
if i < 10: print('')
else: print(i)
# 7. 평가
print('[40] evaluating...')
percentage = round(
100.0 * sumTestThroughput /
sumCorrectMaxThroughput, 6)
print('size:' + str(size) + ' train:' + str(numTrain) +
' test:' + str(numTest) +
' problem(0=Sum,1=Common):' + str(problemNo) +
' epoch:' + str(epoc) + ' dropout:' +
str(round(drop, 3)))
print('total test throughput: ' +
str(round(sumTestThroughput, 6)))
print('total max throughput: ' +
str(round(sumCorrectMaxThroughput, 6)))
print('percentage : ' + str(percentage) +
' %')
RL.append([
size, numTrain, numTest, problemNo, epoc, drop,
percentage
])
# 오류 발생 처리
except Exception as ex:
print('\n################################')
print('error: ' + str(ex))
print('retry training and testing')
print('################################\n')
# 현재까지의 평가 결과 출력
print('\n <<< ESTIMATION RESULT >>>')
print(
'No.\tsize\ttrain\ttest\tproblem\tepoch\tdropout\tpercentage'
)
saveResult = '\n <<< ESTIMATION RESULT >>>\nNo.\tsize\ttrain\ttest\tproblem\tepoch\tdropout\tpercentage\n'
for i in range(len(RL)):
toPrint = (str(i) + '\t' + str(RL[i][0]) + '\t' +
str(RL[i][1]) + '\t' + str(RL[i][2]) +
'\t' + str(RL[i][3]) + '\t' +
str(RL[i][4]) + '\t' +
str(round(RL[i][5], 3)) + '\t' +
(' ' if round(RL[i][6], 4) < 10.0 else '') +
str(round(RL[i][6], 4)))
print(toPrint)
saveResult += (toPrint + '\n')
print('\n')
# 평가 결과 저장
fSave = open(
'DL_WPCN_forPaper_' + str(problemNo) + '_' +
str(size) + '_' + str(numWD) + '.txt', 'w')
fSave.write(saveResult)
fSave.close()
fSave_ = open(
'DL_WPCN_forPaper_' + str(problemNo) + '_' +
str(size) + '_' + str(numWD) + '_detail.txt', 'w')
fSave_.write(resultForThisMap)
fSave_.close()
# 학습 및 테스트 종료
break
# 완전 종료
break
def testNumeric(img, modelName, w, h):
# find height and width of image
width = img.size[0] # width of image
height = img.size[1] # height of image
print('height=' + str(height) + ', width=' + str(width))
# initialize top, bottom, left and right
top = 0
bottom = height
left = width
right = width
recognized = ''
model = DL.deepLearningModel(modelName, True)
lastCheckForCOP = False # latest value of checkForCOP function (init as False)
# crop and recognize number
while left >= 0:
# resize cropped area of image
# resize horizontally -> expand cropped image area to 1 pixels(columns) left
left -= 1
# check if the 'column of pixel' is B-W-B
# not COP, then meet boundary of 'number'
cFCOP = checkForCOP(img, left)
# start of 'number'
if cFCOP == True and lastCheckForCOP == False:
# move right boundary of cropped area to left boundary, and move left boundary 15 pixels left,
# so that cropped area is 15px horizontally
right = left
left = max(left - 15, 1)
# start of boundary of 'number' = end of 'number'
elif cFCOP == False and lastCheckForCOP == True:
# deep learning output for test data (imgArray)
croppedImage = img.crop(
(max(left - 2, 0), top, min(right + 6, width), bottom)) # crop
tempFileName = 'temp.png'
croppedImage.save(
tempFileName) # temporarily save resized image file
croppedImage = Image.open(
tempFileName) # open the temporarily saved file
resizedImage = croppedImage.resize((w, h)) # resize
imArray = np.array(
resizedImage
) # [height][width][3] numpy array of resized image
imArray_ = imArrayToArray(
imArray, True
) # change the shape([height][width][3]) into [height][width]
os.remove(tempFileName) # delete the file
testOutput = DL.modelOutput(model, [imArray_])
# output layer array of test result
outputLayer = testOutput[len(testOutput) - 1][0]
# find index of maximum value in outputLayer
maxIndex = 0 # index of maximum value
maxVal = max(outputLayer)
for i in range(10):
if outputLayer[i] == maxVal: maxIndex = i
# print test result
print('top,bottom,left,right = ' + str(top) + ' ' + str(bottom) +
' ' + str(max(left - 2, 0)) + ' ' +
str(min(right + 6, width)) + ' / maxIndex,maxVal = ' +
str(maxIndex) + ' ' + str(round(maxVal, 6)))
# recognize as number whose match probability is highest
print('recognized as ' + str(maxIndex))
recognized = str(maxIndex) + recognized
# update lastCheckForCOP as current value
lastCheckForCOP = cFCOP
# return result
return recognized
def test(imgArray, answerArray, testFileList, modelName, XAI):
count = len(imgArray) # number of images
correct = 0 # number of correctly classified images
newModel = DL.deepLearningModel(modelName, True)
testOutput = DL.modelOutput(
newModel, imgArray) # deep learning output for test data (imgArray)
# test result
for i in range(count):
outputLayer = testOutput[len(testOutput) -
1][i] # output layer array of test result
# find index of maximum value in outputLayer
maxIndex = 0 # index of maximum value
maxVal = 0
for j in range(10):
if outputLayer[j] > maxVal:
maxIndex = j
maxVal = outputLayer[j]
# compare maxIndex with correct answer
for j in range(10):
outputLayer[j] = round(max(0.000499, outputLayer[j] - 0.0005),
3) # to print
print(testFileList[i] + ' ' * (16 - len(testFileList[i])) +
' / out: ' + str(outputLayer) + ' / ans: ' + str(answerArray[i]))
if answerArray[i][maxIndex] == 1: correct += 1
# using explainable XAI
if XAI == True and random.random(
) < 0.2: # (20% probability for each test image)
thisImg = imgArray[i]
height = len(imgArray[i])
width = len(imgArray[i][0])
# square dif of output layer when changing each pixel by 1 (interval: 2 pixels for both horizontal and vertical)
result = [[0] * int(width) for _ in range(int(height))]
maxSquareDif = 0.0 # max value of square dif (potential)
# modify each pixel by 1 and see the result
for j in range(int(height)):
for k in range(int(width)):
newImg = DLH.arrayCopy(thisImg)
newImg[j][k] += 1
newImgOutput = DL.modelOutput(newModel, [newImg])
# compare output layer
newImgOutputLayer = newImgOutput[len(newImgOutput) - 1][
0] # output layer array of modified image
squareDif = 0
for l in range(10):
squareDif += pow(outputLayer[l] - newImgOutputLayer[l],
2)
# print('modify ' + str(j) + ',' + str(k) + ' by 1 -> square of dif = ' + str(round(squareDif, 8)))
# save into array
result[j][k] = squareDif
if squareDif > maxSquareDif: maxSquareDif = squareDif
# save the square dif array to image
plt.imsave(testFileList[i].split('.')[0] + '_why.png',
np.array(result),
cmap='Greys')
print('correct rate: ' + str(correct) + ' / ' + str(count) + ', ' +
str(round(100 * correct / count, 2)) + '%')