## 1-4. lastScreen, lastDecision, lastPoint 갱신 ##
for j in range(size):
for k in range(size):
lastScreen[j][k] = screen[j][k]
lastDecision = decision
lastPoint = [point[0], point[1]]
### 2. 게임 종료 후 딥러닝 실행 ###
if len(states) < 1000: # 1000개 미만이면 모두 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error', states, outputs, Qdiffs,
25, 'basket', [20, 25], [8, 'sigmoid', 'sigmoid'],
[6, 'sigmoid', 'sigmoid'], False, False, True, deviceName)
else: # 1000개 이상이면 마지막 1000개만 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error',
states[len(states) - 1000:], outputs[len(states) - 1000:],
Qdiffs[len(states) - 1000:], 25, 'basket', [20, 25],
[8, 'sigmoid', 'sigmoid'], [6, 'sigmoid', 'sigmoid'], False,
False, True, deviceName)
isFile = True
if option % 3 == 0 or option % 3 == 1:
model0 = Qlearning_deep_GPU.deepLearningModel('basket_0', False)
if option % 3 == 1 or option % 3 == 2:
model1 = Qlearning_deep_GPU.deepLearningModel('basket_1', False)
print('\n\n << 점수 목록 >>')
print(scores)
print('')
keras.layers.Dropout(drop),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dropout(drop),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dropout(drop),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(2, activation='sigmoid')]
# 5. 옵티마이저
op = tf.keras.optimizers.Adam(0.001)
print('training...')
Qlearning_deep_GPU.deepQlearning(0, NN, op, lc, inputs, outputs, None, None, 'WPCNdeep', [epoc, None], None, None, False, False, True, deviceName)
# 6. 테스트 결과 확인하고 정답과 비교하기
newModel = Qlearning_deep_GPU.deepLearningModel('WPCNdeep_0', False)
sumTestThroughput = 0.0 # test throughput의 합계
sumCorrectMaxThroughput = 0.0 # 정답의 throughput의 합계 (training data를 생성할 때와 같은 방법으로 최대 throughput 확인)
optiInfo = open('optiInfo_' + str(problemNo) + '.txt', 'r')
optiInformation = optiInfo.readlines()
optiInfo.close()
print('testing...')
for i in range(numTest):
testScreen = originalScreen[numTrain + i] # 테스트할 스크린
testOutput = Qlearning_deep_GPU.modelOutput(newModel, [testScreen]) # 테스트 결과
testOutputLayer = testOutput[len(testOutput)-1] # 테스트 결과의 output layer의 값
# 6-0. 테스트 결과 확인
## 1-4. lastScreen, lastDecision, lastPoint 갱신 ##
for j in range(size):
for k in range(size):
lastScreen[j][k] = screen[j][k]
lastDecision = decision
lastPoint = [point[0], point[1]]
### 2. 게임 종료 후 딥러닝 실행 ###
if len(states) < 1000: # 1000개 미만이면 모두 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error', states, outputs, Qdiffs,
25, 'vectorCar', [20, 25], [8, 'sigmoid', 'sigmoid'],
[6, 'sigmoid', 'sigmoid'], False, False, True, deviceName)
else: # 1000개 이상이면 마지막 1000개만 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error',
states[len(states) - 1000:], outputs[len(states) - 1000:],
Qdiffs[len(states) - 1000:], 25, 'vectorCar', [20, 25],
[8, 'sigmoid', 'sigmoid'], [6, 'sigmoid', 'sigmoid'], False,
False, True, deviceName)
isFile = True
if option % 3 == 0 or option % 3 == 1:
model0 = Qlearning_deep_GPU.deepLearningModel('vectorCar_0', False)
if option % 3 == 1 or option % 3 == 2:
model1 = Qlearning_deep_GPU.deepLearningModel('vectorCar_1', False)
print('\n\n << 점수 목록 >>')
print(scores)
print('')
## 1-4. lastScreen, lastDecision, lastPoint 갱신 ##
for j in range(size):
for k in range(size):
lastScreen[j][k] = screen[j][k]
lastDecision = decision
lastPoint = [point[0], point[1]]
### 2. 게임 종료 후 딥러닝 실행 ###
if len(states) < 1000: # 1000개 미만이면 모두 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error', states, outputs, Qdiffs,
25, 'valueMaze', [20, 25], [8, 'sigmoid', 'sigmoid'],
[6, 'sigmoid', 'sigmoid'], False, False, True, deviceName)
else: # 1000개 이상이면 마지막 1000개만 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error',
states[len(states) - 1000:], outputs[len(states) - 1000:],
Qdiffs[len(states) - 1000:], 25, 'valueMaze', [20, 25],
[8, 'sigmoid', 'sigmoid'], [6, 'sigmoid', 'sigmoid'], False,
False, True, deviceName)
isFile = True
if option % 3 == 0 or option % 3 == 1:
model0 = Qlearning_deep_GPU.deepLearningModel('valueMaze_0', False)
if option % 3 == 1 or option % 3 == 2:
model1 = Qlearning_deep_GPU.deepLearningModel('valueMaze_1', False)
print('\n\n << 점수 목록 >>')
print(scores)
print('')
keras.layers.Dropout(drop),
keras.layers.Dense(40, activation='elu'),
keras.layers.Dense(2, activation='sigmoid')]
# 5. 옵티마이저
op0 = tf.keras.optimizers.Adam(0.001)
op1 = tf.keras.optimizers.Adam(0.001)
op2 = tf.keras.optimizers.Adam(0.001)
print('training...')
Qlearning_deep_GPU.deepQlearning(0, NN2, op0, lc, inputs, outputs, None, None, 'WPCNdeepNN2', [epoc, None], None, None, False, False, True, deviceName)
Qlearning_deep_GPU.deepQlearning(0, NN1, op1, lc, inputs, outputs, None, None, 'WPCNdeepNN1', [epoc, None], None, None, False, False, True, deviceName)
Qlearning_deep_GPU.deepQlearning(0, NN0, op2, lc, inputs, outputs, None, None, 'WPCNdeepNN0', [epoc, None], None, None, False, False, True, deviceName)
# 6. 테스트 결과 확인하고 정답과 비교하기
newModel0 = Qlearning_deep_GPU.deepLearningModel('WPCNdeepNN0_0', False)
newModel1 = Qlearning_deep_GPU.deepLearningModel('WPCNdeepNN1_0', False)
newModel2 = Qlearning_deep_GPU.deepLearningModel('WPCNdeepNN2_0', False)
sumTestThroughput = 0.0 # test throughput의 합계
sumCorrectMaxThroughput = 0.0 # 정답의 throughput의 합계 (training data를 생성할 때와 같은 방법으로 최대 throughput 확인)
optiInfo = open('optiInfo_' + str(problemNo) + '.txt', 'r')
optiInformation = optiInfo.readlines()
optiInfo.close()
print('testing...')
for i in range(numTest):
testScreen = originalScreen[numTrain + i] # 테스트할 스크린
## 1-4. lastScreen, lastDecision, lastPoint 갱신 ##
for j in range(size):
for k in range(size):
lastScreen[j][k] = screen[j][k]
lastDecision = decision
lastPoint = [point[0], point[1]]
### 2. 게임 종료 후 딥러닝 실행 ###
if len(states) < 1000: # 1000개 미만이면 모두 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error', states, outputs, Qdiffs,
25, 'snake', [20, 25], [8, 'sigmoid', 'sigmoid'],
[6, 'sigmoid', 'sigmoid'], False, False, True, deviceName)
else: # 1000개 이상이면 마지막 1000개만 학습
Qlearning_deep_GPU.deepQlearning(
option, NN, op, 'mean_squared_error',
states[len(states) - 1000:], outputs[len(states) - 1000:],
Qdiffs[len(states) - 1000:], 25, 'valueMaze', [20, 25],
[8, 'sigmoid', 'sigmoid'], [6, 'sigmoid', 'sigmoid'], False,
False, True, deviceName)
isFile = True
if option % 3 == 0 or option % 3 == 1:
model0 = Qlearning_deep_GPU.deepLearningModel('snake_0', False)
if option % 3 == 1 or option % 3 == 2:
model1 = Qlearning_deep_GPU.deepLearningModel('snake_1', False)
print('\n\n << 점수 목록 >>')
print(scores)
print('')
print('\n ' + ('#===' * 16) + ' <<<< option:' + str(i) +
' LEARNING >>>> ' + ('===#' * 16) + '\n')
Qlearning_deep_GPU.deepQlearning(i, NN, op, 'mean_squared_error',
states, outputs, Qdiff, 25,
'test' + str(i), [20, 25],
[8, 'sigmoid', 'sigmoid'],
[6, 'sigmoid', 'sigmoid'], True, True,
False, deviceName)
# test
print('\n ' + ('#===' * 16) + ' <<<< option:' + str(i) +
' TEST >>>> ' + ('===#' * 16) + '\n')
if i % 3 == 0 or i % 3 == 1:
print('\n << test output (첫번째 Neural Network) >>\n')
newModel = Qlearning_deep_GPU.deepLearningModel(
'test' + str(i) + '_0', False)
testOutput = Qlearning_deep_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 j in range(len(testOutput)):
print(' << layer No: ' + str(j) + ' >>')
print(str(testOutput[j]))
if i % 3 == 1 or i % 3 == 2:
print('\n << test output (두번째(또는 Dueling) Neural Network) >>\n')
newModel = Qlearning_deep_GPU.deepLearningModel(
'test' + str(i) + '_1', False)
testOutput = Qlearning_deep_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')