def generateExamples(dataset, predClasses):
'''Given the entire dataset and a predicted label, print TOP_EG number of example pairs.'''
# Init
labels = [dataset.dict_Index_Cluster[i] for i in predClasses]
prettyLabels = [i.replace('\n', ' ') for i in labels]
examplesList = []
print 'Top-{} Predicted Class-IDs:\n{}'.format(TOP_K, predClasses)
print 'TOP-{} Predicted Class-Labels (Error-IDs and Unique Repairs):\n{}\nExamples:'.format(TOP_K, prettyLabels)
# For each label
numC = 0
for label in labels:
dataSamples = dataset.dict_ClusterRaw_DataSamples[label] # Fetch all training examples
srcTrgtPairs = [(H.joinList(i.src, ''), H.joinList(i.trgt, '')) for i in dataSamples] # Create list of (before, after)
topEgs = H.getTop_K(srcTrgtPairs, TOP_EG) # Sort them based on frequency and fetch Top-EG #examples
for i in range(len(topEgs)): # For each example
eg = topEgs[i]
before, after = eg[0].replace('\n', ''), eg[1].replace('\n', '') # Remove new-line characters
if before.strip() == '': before = '// Empty Line' # Deal with empty lines
if after.strip() == '': after = '// Empty Line' # Deal with empty lines
print 'Eg #{} class-{} before: {}'.format(i+1, predClasses[numC], before) # Print them
print 'Eg #{} class-{} after : {}'.format(i+1, predClasses[numC], after)
examplesList.append(eg)
numC += 1
# Extend to the examplesList, proportionately
return examplesList
def getBuggyAbsLine(codeText):
'''Given codeText, return the buggy abstract lines (abstraction of erroneous lines) and their line numbers'''
codeObj = Code(codeText)
absLines = getProgAbstraction(codeObj)
errs = codeObj.getSevereErrors()
absLinesBuggy, lineNums = [], []
if len(errs) > 0:
for err in errs:
lineNum = err.line # Pick the first error line Num
if lineNum > 0 and lineNum <= len(absLines):
# If line-num reported by compiler doesn't exceeds #absLines (and is >=1)
absLine = H.joinList(absLines[lineNum - 1], ' ')
if lineNum not in lineNums: # Add unique lineNum / absLine
absLinesBuggy.append(absLine)
lineNums.append(lineNum)
return absLinesBuggy, lineNums
def printProgAbstraction(fnamePath):
codeText = open(fnamePath).read()
codeObj = Code(codeText)
absLines = getProgAbstraction(codeObj)
for line in absLines:
print H.joinList(line, ' ')
def calcConfMat(self):
print colored('\n\tConfusion Matrix: ...', 'magenta')
predAtK = []
for topK in [1, 3, 5]:
countM, countN = 0, 0
predClasses_Tests = self.deepModel.getPrediction(topK)
for actClasses_bin, predClasses_bin in zip(self.y_test,
predClasses_Tests):
if self.multiClass:
# Add +1 to index since off-by-one with dict_index2class
actClasses_indices = [
index + 1 for index in range(len(actClasses_bin))
if actClasses_bin[index] == 1
]
predClasses_indices = [
index + 1 for index in range(len(predClasses_bin))
if predClasses_bin[index] == 1
]
else:
actClasses_indices, predClasses_indices = [
np.argmax(actClasses_bin)
], predClasses_bin
actClasses = [
self.dict_index2class[index]
for index in actClasses_indices
]
predClasses = [
self.dict_index2class[index]
for index in predClasses_indices
]
for actClass in actClasses:
if actClass in predClasses: # True-Positive
countM += 1
if topK == 1: # Conf Matrix only for Pred@1
self.confMatrix[actClass].truePos += 1
self.confMatrix[actClass].updatePred(actClass)
else: # False-Negative: Not predicted at all
countN += 1
if topK == 1: # Conf Matrix only for Pred@1
self.confMatrix[actClass].falseNeg += 1
for predClass in predClasses: # Add all confusion labels
self.confMatrix[actClass].updatePred(predClass)
if topK == 1: # Conf Matrix only for Pred@1
for predClass in predClasses:
if predClass not in actClasses: # False-Positive: Predicted, but falsely
self.confMatrix[predClass].falsePos += 1
if topK == 1: # Conf Matrix only for Pred@1
sortedConfMat = [
self.confMatrix[i] for i in self.getSortedConfMat()[0]
]
strConf = H.joinList(sortedConfMat)
self.writeConfMat(sortedConfMat)
prec_at_k = round(100 * float(countM) / (countM + countN), 2)
predAtK.append(prec_at_k)
print 'Pred@{}= {}'.format(topK, prec_at_k)
return strConf, predAtK
def __str__(self):
return H.joinList(self.abstractTokens, ' ')