def pintoKNN(outDir, projectBasePath, projectName, kf):
v0 = time.perf_counter()
dataPointsFlaky, dataPointsNonFlaky = flast.getDataPointsInfo(
projectBasePath, projectName)
dataPoints = dataPointsFlaky + dataPointsNonFlaky
countVec = CountVectorizer()
Z = countVec.fit_transform(dataPoints)
dataPointsList = np.array([Z[i].toarray() for i in range(Z.shape[0])])
dataLabelsList = np.array([1] * len(dataPointsFlaky) +
[0] * len(dataPointsNonFlaky))
v1 = time.perf_counter()
vecTime = v1 - v0
# storage
kNN = (dataPointsList, dataLabelsList)
pickleDumpKNN = os.path.join(outDir, "PintoKNN.pickle")
with open(pickleDumpKNN, "wb") as pickleFile:
pickle.dump(kNN, pickleFile)
storage = os.path.getsize(pickleDumpKNN)
os.remove(pickleDumpKNN)
avgP, avgR = 0, 0
avgTPrep, avgTPred = 0, 0
successFold, precisionFold = 0, 0
for (trnIdx, tstIdx) in kf.split(dataPointsList, dataLabelsList):
trainData, testData = dataPointsList[trnIdx], dataPointsList[tstIdx]
trainLabels, testLabels = dataLabelsList[trnIdx], dataLabelsList[
tstIdx]
if sum(trainLabels) == 0 or sum(testLabels) == 0:
print("Skipping fold...")
print(" Flaky Train Tests", sum(trainLabels))
print(" Flaky Test Tests", sum(testLabels))
continue
successFold += 1
# prepare the data in the right format for kNN
nSamplesTrainData, nxTrain, nyTrain = trainData.shape
trainData = trainData.reshape((nSamplesTrainData, nxTrain * nyTrain))
nSamplesTestData, nxTest, nyTest = testData.shape
testData = testData.reshape((nSamplesTestData, nxTest * nyTest))
# training
t0 = time.perf_counter()
kNN = KNeighborsClassifier(algorithm="brute",
metric="euclidean",
weights="uniform",
n_neighbors=1,
n_jobs=1)
kNN.fit(trainData, trainLabels)
t1 = time.perf_counter()
trainTime = t1 - t0
# testing
p0 = time.perf_counter()
predictLabels = kNN.predict(testData)
p1 = time.perf_counter()
testTime = p1 - p0
preparationTime = (vecTime * len(trainData) /
len(dataPoints)) + trainTime
predictionTime = (vecTime / len(dataPoints)) + (testTime /
len(testData))
(precision, recall) = flast.computeResults(testLabels, predictLabels)
print(precision, recall)
if precision != "-":
precisionFold += 1
avgP += precision
avgR += recall
avgTPrep += preparationTime
avgTPred += predictionTime
if precisionFold == 0:
avgP = "-"
else:
avgP /= precisionFold
avgR /= successFold
avgTPrep /= successFold
avgTPred /= successFold
return (avgP, avgR, storage, avgTPrep, avgTPred)
for projectName in projectList:
for threshold in [0.5, 0.95]:
p0 = time.time()
print("#" * 80)
print("TESTING {}".format(projectName.upper()))
pName = projectName.split("/")[-1]
outputFile = f"{outDir}/{pName}___t{threshold}.csv"
with open(outputFile, "w") as fileOut:
fileOut.write(
"fold,testSetSize,numFlakyTrainSet,numNonFlakyTrainSet,numFlakyTestSet,numNonFlakyTestSet,vecTime,trainTime,testTime,avgPredTime,f-measure,precision,recall,accuracy,tp,fp,fn,tn\n"
)
# data points vectorization
v0 = time.perf_counter()
dataPointsFlaky, dataPointsNonFlaky = flast.getDataPointsInfo(
projectBasePath, projectName)
dataPoints = dataPointsFlaky + dataPointsNonFlaky
Z = flast.flastVectorization(dataPoints,
reduceDim=reduceDim,
dim=dim,
eps=eps)
dataPointsList = np.array(
[Z[i].toarray() for i in range(Z.shape[0])])
dataLabelsList = np.array([1] * len(dataPointsFlaky) +
[0] * len(dataPointsNonFlaky))
v1 = time.perf_counter()
vecTime = v1 - v0
for testSetSize in testSetSizes:
t0 = time.time()
print()
def flastKNN(outDir, projectBasePath, projectName, kf, dim, eps, k, sigma,
params):
v0 = time.perf_counter()
dataPointsFlaky, dataPointsNonFlaky = flast.getDataPointsInfo(
projectBasePath, projectName)
dataPoints = dataPointsFlaky + dataPointsNonFlaky
Z = flast.flastVectorization(dataPoints, dim=dim, eps=eps)
dataPointsList = np.array([Z[i].toarray() for i in range(Z.shape[0])])
dataLabelsList = np.array([1] * len(dataPointsFlaky) +
[0] * len(dataPointsNonFlaky))
v1 = time.perf_counter()
vecTime = v1 - v0
# storage
kNN = (dataPointsList, dataLabelsList)
pickleDumpKNN = os.path.join(outDir,
"flast-k{}-sigma{}.pickle".format(k, sigma))
with open(pickleDumpKNN, "wb") as pickleFile:
pickle.dump(kNN, pickleFile)
storage = os.path.getsize(pickleDumpKNN)
os.remove(pickleDumpKNN)
avgP, avgR = 0, 0
avgTPrep, avgTPred = 0, 0
successFold, precisionFold = 0, 0
for (trnIdx, tstIdx) in kf.split(dataPointsList, dataLabelsList):
trainData, testData = dataPointsList[trnIdx], dataPointsList[tstIdx]
trainLabels, testLabels = dataLabelsList[trnIdx], dataLabelsList[
tstIdx]
if sum(trainLabels) == 0 or sum(testLabels) == 0:
print("Skipping fold...")
print(" Flaky Train Tests", sum(trainLabels))
print(" Flaky Test Tests", sum(testLabels))
continue
successFold += 1
# prepare the data in the right format for kNN
nSamplesTrainData, nxTrain, nyTrain = trainData.shape
trainData = trainData.reshape((nSamplesTrainData, nxTrain * nyTrain))
nSamplesTestData, nxTest, nyTest = testData.shape
testData = testData.reshape((nSamplesTestData, nxTest * nyTest))
trainTime, testTime, predictLabels = flast.flastClassification(
trainData, trainLabels, testData, sigma, k, params)
preparationTime = (vecTime * len(trainData) /
len(dataPoints)) + trainTime
predictionTime = (vecTime / len(dataPoints)) + (testTime /
len(testData))
(precision, recall) = flast.computeResults(testLabels, predictLabels)
print(precision, recall)
if precision != "-":
precisionFold += 1
avgP += precision
avgR += recall
avgTPrep += preparationTime
avgTPred += predictionTime
if precisionFold == 0:
avgP = "-"
else:
avgP /= precisionFold
avgR /= successFold
avgTPrep /= successFold
avgTPred /= successFold
return (avgP, avgR, storage, avgTPrep, avgTPred)