def train(self):
allSamples = goodSamples = 0.0
self.log('Starting the real training...', 'main')
if self.sender() == self.ui.actionTrain:
procData = processData(self.trainPat)
result = procData.readContents()
data = inputData()
data.addAll(result)
self.ls = Lamstar.lamstar(15, 1)
for iter in range(10):
print('Iteration %s' % iter)
for i in range(data.getCount()):
print('Training data : ', i)
self.arr2Image(data.getWholeArray(i), 'input')
#raw_input('Enter')
self.ls.train(data.getSubWords(i), data.getOutputs(i))
if (self.debug > 0):
self.log('Iteration no:' + str(iter), 'main')
self.log('No of nodes' + str(self.ls.getNoOfNodes()),
'main')
elif self.sender() == self.ui.actionRun_Test:
#goodSamples = allSamples = 0
procData = processData(self.testPat)
result = procData.readContents()
data = inputData()
data.addAll(result)
files = []
for afile in os.listdir(self.testingDir):
if afile[-3:] == 'png' and afile.find('procced') != -1:
files.append(afile)
files.sort()
print('Quering for test data')
self.log('Quering for test data', 'main')
print('number of items: %s' % (data.getCount()))
for i in range(data.getCount()):
self.log('Querying for ' + files[i].rpartition('/')[-1],
'query')
print('Querying for ' + files[i].rpartition('/')[-1])
self.arr2Image(data.getWholeArray(i), 'input')
(result, BMU) = self.ls.query(data.getSubWords(i))
arr = self.constructArr(BMU)
self.arr2Image(arr, 'output')
self.log(result, 'status')
allSamples += 1
if files[i].find(result + '-') != -1:
goodSamples += 1
self.ui.resultLabel.setText(result)
self.log('No of nodes: ' + str(self.ls.getNoOfNodes()), 'main')
self.log('No of links: ' + str(self.ls.getNoOfLinks()), 'main')
self.log(
'Accuracy = ' + str(
(float(goodSamples) / allSamples) * 100) + '%', 'main')
print('Good sample=%s all=%s' % (goodSamples, allSamples))
def train(self):
allSamples = goodSamples = 0.0
self.log('Starting the real training...', 'main')
if self.sender() == self.ui.actionTrain:
procData = processData(self.trainPat)
result = procData.readContents()
data = inputData()
data.addAll(result)
self.ls = Lamstar.lamstar(15, 1)
for iter in range(10):
print('Iteration %s' % iter)
for i in range(data.getCount()):
# print('Training data : ', i)
# self.arr2Image(data.getWholeArray(i),'input')
# raw_input('Enter')
self.ls.train(data.getSubWords(i), data.getOutputs(i))
if(self.debug > 0):
self.log('Iteration no:' + str(iter), 'main')
self.log('No of nodes' + str(self.ls.getNoOfNodes()), 'main')
elif self.sender() == self.ui.actionRun_Test:
goodSamples = allSamples = 0
procData = processData(self.testPat)
result = procData.readContents()
data = inputData()
data.addAll(result)
files = []
for afile in os.listdir(self.testingDir):
if afile[-3:] == 'jpg' and afile.find('procced') != -1:
files.append(afile)
files.sort()
print('Quering for test data')
self.log('Quering for test data', 'main')
print('number of items: %s' % (data.getCount()))
for i in range(data.getCount()):
self.log('Querying for ' + files[i].rpartition('/')[-1], 'query')
print('Querying for ' + files[i].rpartition('/')[-1])
self.arr2Image(data.getWholeArray(i), 'input')
(result, BMU) = self.ls.query(data.getSubWords(i))
arr = self.constructArr(BMU)
self.arr2Image(arr, 'output')
self.log(result, 'status')
allSamples += 1
if files[i].find(result + '-') != -1:
goodSamples += 1
self.ui.resultLabel.setText(result)
self.log('No of nodes: ' + str(self.ls.getNoOfNodes()), 'main')
self.log('No of links: ' + str(self.ls.getNoOfLinks()), 'main')
self.log('Accuracy = ' + str((float(goodSamples) / allSamples) * 100) + '%', 'main')
print ('Good sample=%s all=%s' % (goodSamples, allSamples))
def getNoOfLinks(self):
return len(self.Link2OutputDict)
def printTable(self):
for key in sorted(self.LinkTable.keys()):
print('%s %s' % (key, self.LinkTable[key]))
def maxFromDict(self, dictionary):
return max(dictionary.iteritems(), key=lambda item: sum(item[1]))[0]
if __name__ == '__main__':
procData = processData('data/images/processed/all/150x200.pat')
result = procData.readContents()
data = inputData()
data.addAll(result)
preData = processData('data/images/processed/unseen/unseen.pat')
result = preData.readContents()
unseenData = inputData()
unseenData.addAll(result)
ls = lamstar(10, 1)
for _ in range(10):
for i in range(data.getCount()):
# print('Training data : ', i)
ls.train(data.getSubWords(i), data.getOutputs(i))
# sys.stdin.read(1)
# ls.printSOMs()
def printTable(self):
for key in self.LinkTable.keys():
print('%s %s' % (key, self.LinkTable[key]))
def maxFromDict(self, dictionary):
#return min(dictionary.items(), key=lambda item: sum(item[1])/float(len(item[1])))[0]
return max(dictionary.items(), key=lambda item: sum(item[1]))[0]
if __name__ == '__main__':
#procData = processData('data/images/processed/all/150x200.pat')
procData = processData('data/clevepro.pat')
result = procData.readContents()
data = inputData()
data.addAll(result)
# #preData = processData('data/images/processed/unseen/unseen.pat')
# preData = processData('data/.txt')
# result = preData.readContents()
# unseenData = inputData()
# unseenData.addAll(result)
ls = lamstar(10, 1)
for _ in range(10):
for i in range(data.getCount()):
print('Training data : ', i)
ls.train(data.getSubWords(i), data.getOutputs(i))
#sys.stdin.read(1)
def train(
self,
traindata,
iternum,
distth,
bywho,
):
procData = processData(traindata, self.label)
result = procData.readContents()
data = inputData()
data.addAll(result)
b = []
for i in range(0, data.getCount()):
b.append(i)
if bywho == 'train':
if self.trainnumber > data.getCount(): #에외처리
print(
"training data is larger than number of dataset. Choose trainnum lower than "
+ str(data.getCount()))
return 0
self.ls = Lamstar.lamstar(self.subwordnumber,
1) #램스타 객체 생성(서브워드 갯수만큼)
#b = random.sample(range(0, data.getCount()), self.trainnumber)
for iter in range(iternum):
#print('Training Iteration %s' % iter)
for i in range(self.trainnumber): #num of subwords
#print('Training data : ', b[i]) #샘플링 데이터 인덱스 확인 가능
self.ls.train(data.getSubWords(b[i], self.subwordlist),
data.getOutputs(b[i]), distth) #트레이닝
elif bywho == 'test':
b = b[data.getCount() - self.testnumber:data.getCount()]
accuarcy = 0
TP = 0
TN = 0
FP = 0
FN = 0
#if self.testnumber > data.getCount():
# print("testing data is larger than number of dataset. Choose testnum lower than " + str(data.getCount()))
# return 0
print("Training iteration: " + str(self.iternum))
#a =[]
#for i in range(0, data.getCount()):
# a.append(i)
for i in range(len(b)):
# p = a[random.randrange(0, len(a) - 1)]
#print("")
#print("Validation "+str(i))
queryvalue = self.ls.query(
data.getSubWords(b[i], self.subwordlist))
if queryvalue == data.getOutputs(b[i]):
# print("GT : " + data.getOutputs(i) +" ST : " + queryvalue + " Result : TRUE")
accuarcy += 1
# else:
# print("GT : " + data.getOutputs(i) + " ST : " + queryvalue + " Result : FALSE")
# print("")
if int(data.getOutputs(b[i])) == 0 and int(queryvalue) == 0:
TN += 1
if int(data.getOutputs(b[i])) == 1 and int(queryvalue) == 1:
TP += 1
if int(data.getOutputs(b[i])) == 0 and int(queryvalue) == 1:
FP += 1
if int(data.getOutputs(b[i])) == 1 and int(queryvalue) == 0:
FN += 1
accuarcy /= int(len(b)) # 분할 validation 버젼
#accuarcy /= self.testnumber #랜덤 샘플링 버젼
#print("Accuarcy : " + str(accuarcy * 100) + "%")
if TP == 20: exit
else:
if TP + FP == 0:
prec = 0
elif TN + FP == 0:
print("TN + FP is Zero! setting prec to 0")
prec = 0
else:
prec = TN / (TN + FP)
if TP + FN == 0:
rec = 0
else:
rec = TP / (TP + FN)
#accuracy = (TP + TN)/(TP + TN + FP + FN)
#if (rec == 0) and (prec == 0): exit
print("Sensitivity:" + str(rec * 100) + "%",
"Precision:" + str(prec * 100) + "%",
"Accuarcy:" + str(accuarcy * 100) + "%")
print("TP:" + str(TP) + " TN:" + str(TN) + " FP:" + str(FP) +
" FN:" + str(FN))
print("")
