def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
# deep learning
if config.DEEP_LEARNING_METHOD != -1:
if config.DEEP_LEARNING_METHOD == 1:
logistic_sgd_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 2:
[trainingFile,
testingFile] = dA_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 3:
mlp_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 4:
SdA_2.runDL([trainingFile, testingFile])
if config.NUM_MONITORED_SITES != -1:
[accuracy, debugInfo] = ['NA', []]
return [accuracy, debugInfo]
'''
return wekaAPI.execute( trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
['-K','2', # RBF kernel
'-G','0.0000019073486328125', # Gamma
'-C','131072'] ) # Cost
'''
if config.CROSS_VALIDATION == 0:
return wekaAPI.execute(
trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072'
]) # Cost
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072'
]) # Cost
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
if config.n_components_PCA != 0:
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
if config.n_components_LDA != 0:
[trainingFile,
testingFile] = Utils.calcLDA6([trainingFile, testingFile])
if config.n_components_QDA != 0:
[trainingFile,
testingFile] = Utils.calcQDA([trainingFile, testingFile])
classifier = "svm"
kwargs = {}
kwargs['C'] = 2**11
kwargs['kernel'] = 'rbf'
kwargs['gamma'] = 2
if config.CROSS_VALIDATION == 0:
return wekaAPI.executeSklearn(trainingFile, testingFile,
classifier, **kwargs)
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeSklearnCrossValidationScaleWithRange(
file, classifier, config.CROSS_VALIDATION, (-1, 1),
**kwargs) # CV with normalization
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
return wekaAPI.execute(trainingFile, testingFile,
"weka.classifiers.bayes.NaiveBayesMultinomial",
[])
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
if (config.CLUSTERING_METHOD == 1):
clusteringAPI.calcKmeans([trainingFile, testingFile],
"Description goes here!")
elif (config.CLUSTERING_METHOD == 2):
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
clusteringAPI.calcKmeans([trainingFile, testingFile],
"Description goes here!")
elif (config.CLUSTERING_METHOD == 3):
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
clusteringAPI.calcKmeansCvxHullDelaunay(
[trainingFile, testingFile], "Description goes here!")
elif (config.CLUSTERING_METHOD == 4):
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
clusteringAPI.calcKmeansCvxHullDelaunay_Mixed(
[trainingFile, testingFile], "Description goes here!")
elif (config.CLUSTERING_METHOD == 5):
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
clusteringAPI.calcKmeansCvxHullDelaunay_Mixed_KNN(
[trainingFile, testingFile],
"Description goes here!",
threshold=3)
return ['NA', []]
def classify( runID, trainingSet, testingSet ):
[trainingFile,testingFile] = arffWriter.writeArffFiles( runID, trainingSet, testingSet )
return wekaAPI.execute( trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
['-K','2', # RBF kernel
'-G','0.0000019073486328125', # Gamma
'-C','131072'] ) # Cost
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
# deep learning (AE)
if config.AE != -1:
[trainingFile,
testingFile] = dA_2.calcAE([trainingFile, testingFile])
return wekaAPI.execute(trainingFile, testingFile,
"weka.classifiers.bayes.NaiveBayes", ['-K'])
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
if config.n_components_PCA != 0:
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
if config.n_components_LDA != 0:
[trainingFile,
testingFile] = Utils.calcLDA6([trainingFile, testingFile])
if config.n_components_QDA != 0:
[trainingFile,
testingFile] = Utils.calcQDA([trainingFile, testingFile])
return wekaAPI.execute(trainingFile, testingFile,
"weka.classifiers.bayes.NaiveBayes", ['-K'])
def classify( runID, trainingSet, testingSet ):
[trainingFile,testingFile] = arffWriter.writeArffFiles( runID, trainingSet, testingSet )
if config.n_components_PCA != 0:
[trainingFile,testingFile] = Utils.calcPCA2([trainingFile,testingFile])
if config.n_components_LDA != 0:
[trainingFile,testingFile] = Utils.calcLDA6([trainingFile,testingFile])
if config.n_components_QDA != 0:
[trainingFile,testingFile] = Utils.calcQDA([trainingFile,testingFile])
return wekaAPI.execute( trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
['-K','2', # RBF kernel
'-G','0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C','131072'] ) # Cost
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
# return wekaAPI.execute( trainingFile, testingFile, "weka.classifiers.bayes.NaiveBayes", ['-K'] )
if config.CROSS_VALIDATION == 0:
return wekaAPI.execute(trainingFile, testingFile,
"weka.classifiers.bayes.NaiveBayes", ['-K'])
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.classifiers.bayes.NaiveBayes",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K'
])
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
# return wekaAPI.execute( trainingFile, testingFile, "weka.classifiers.bayes.NaiveBayes", ['-K'] )
# deep learning
if config.DEEP_LEARNING_METHOD != -1:
#DLMethod = Utils.intToDL(config.DEEP_LEARNING_METHOD)
#print 'Deep Learning Method: ' + DLMethod
#[trainingFile, testingFile] = DLMethod.runDL([trainingFile, testingFile])
#[trainingFile, testingFile] = dA_2.calcAE([trainingFile, testingFile])
#SdA_2.calcSdA([trainingFile, testingFile])
#logistic_sgd_2.calcLog_sgd([trainingFile, testingFile])
if config.DEEP_LEARNING_METHOD == 1:
logistic_sgd_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 2:
[trainingFile,
testingFile] = dA_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 3:
mlp_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 4:
SdA_2.runDL([trainingFile, testingFile])
if config.CROSS_VALIDATION == 0:
return wekaAPI.execute(trainingFile, testingFile,
"weka.classifiers.bayes.NaiveBayes", ['-K'])
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.classifiers.bayes.NaiveBayes",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K'
])
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
# deep learning (AE)
if config.DEEP_LEARNING_METHOD != -1:
#[trainingFile, testingFile] = dA_2.calcAE([trainingFile, testingFile]) # one layer dA
#[trainingFile, testingFile] = dA_2.calcAE([trainingFile, testingFile]) # two layers dA
#[trainingFile, testingFile] = dA_2.calcAE([trainingFile, testingFile])
#SdA_2.calcSdA([trainingFile, testingFile])
if config.DEEP_LEARNING_METHOD == 1:
[trainingFile, testingFile
] = logistic_sgd_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 2:
[trainingFile,
testingFile] = dA_2.runDL([trainingFile, testingFile])
[trainingFile,
testingFile] = dA_2.runDL([trainingFile, testingFile])
#[trainingFile, testingFile] = dA_2.runDL([trainingFile, testingFile])
#[trainingFile, testingFile] = dA_2.runDL([trainingFile, testingFile])
#[trainingFile, testingFile] = dA_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 3:
# DL classifier
return mlp_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 4:
return SdA_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 5:
return mlp_3.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 6:
return SdA_3.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 7:
return LeNetConvPoolLayer_2.runDL([trainingFile, testingFile])
return wekaAPI.execute(trainingFile, testingFile,
"weka.classifiers.bayes.NaiveBayes", ['-K'])
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
if config.NUM_MONITORED_SITES != -1: #no need to classify as this is for generating openworld datasets. See the line above (arffWriter)
[accuracy, debugInfo] = ['NA', []]
return [accuracy, debugInfo]
if config.n_components_PCA != 0:
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
if config.n_components_LDA != 0:
[trainingFile,
testingFile] = Utils.calcLDA4([trainingFile, testingFile])
if config.n_components_QDA != 0:
[trainingFile,
testingFile] = Utils.calcQDA([trainingFile, testingFile])
if config.lasso != 0:
#[trainingFile,testingFile] = Utils.calcLasso3([trainingFile,testingFile])
#[trainingFile,testingFile] = Utils.calcLogisticRegression([trainingFile,testingFile])
Utils.calcLogisticRegression([trainingFile, testingFile])
#Utils.plotDensity([trainingFile,testingFile])
#Utils.plot([trainingFile,testingFile])
'''
if (config.DATA_SOURCE == 62 or config.DATA_SOURCE == 63 or config.DATA_SOURCE == 64 or config.DATA_SOURCE == 65):
if config.LABEL_NOISE_RATIO != 0:
[trainingFile,testingFile] = Utils.makeLabelNoise([trainingFile,testingFile],config.LABEL_NOISE_RATIO)
'''
if config.CROSS_VALIDATION == 0:
return wekaAPI.execute(
trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
#'-S','2', # one-class svm
'-B'
]) # confidence
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
'-B'
]) # confidence
def classify( runID, trainingSet, testingSet ):
[trainingFile,testingFile] = arffWriter.writeArffFiles( runID, trainingSet, testingSet )
return wekaAPI.execute( trainingFile, testingFile, "weka.classifiers.bayes.NaiveBayes", ['-K'] )
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
if config.n_components_PCA != 0:
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
if config.n_components_LDA != 0:
[trainingFile,
testingFile] = Utils.calcLDA6([trainingFile, testingFile])
if config.n_components_QDA != 0:
[trainingFile,
testingFile] = Utils.calcQDA([trainingFile, testingFile])
classifier = "RF"
kwargs = {}
kwargs[
'n_estimators'] = 500 #number of trees/ length of the fingerprint
kwargs['criterion'] = "gini"
kwargs['oob_score'] = True
kwargs['n_jobs'] = 3
if config.NUM_MONITORED_SITES == -1 and config.NUM_NON_MONITORED_SITES == -1:
# closed world
if config.CROSS_VALIDATION == 0:
return wekaAPI.executeSklearn(trainingFile, testingFile,
classifier, **kwargs)
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeSklearnCrossValidation(
file, classifier, config.CROSS_VALIDATION,
**kwargs) # CV with normalization
else:
# open world
trainList = wekaAPI.readFile(trainingFile)
testList = wekaAPI.readFile(testingFile)
trainInstancesList = []
testInstancesList = []
classes = ""
yTrain = []
yTest = []
for line in trainList:
if line[0] == '@':
if line.lower().startswith("@attribute class"):
classes = line.split(" ")[2]
else:
# instancesList.append(float(line.split(",")[:-1]))
trainInstancesList.append(
[float(i) for i in line.split(",")[:-1]])
yTrain.append(line.split(",")[-1])
for line in testList:
if line[0] != '@':
testInstancesList.append(
[float(i) for i in line.split(",")[:-1]])
yTest.append(line.split(",")[-1])
XTr = numpy.array(trainInstancesList)
yTr = numpy.array(yTrain)
XTe = numpy.array(testInstancesList)
yTe = numpy.array(yTest)
clf = RandomForestClassifier(**kwargs)
print('Generating Leaves...')
training_leaves = clf.apply(XTr)
test_leaves = clf.apply(XTe)
training_leaves = [
numpy.array(training_leaf, dtype=int)
for training_leaf in training_leaves
]
test_leaves = [
numpy.array(test_leaf, dtype=int) for test_leaf in test_leaves
]
true_positive = 0
false_positive = 0
knn = 3 # k value
debugInfo = []
print('Calculating Distances...')
for test_leaf_idx in range(len(test_leaves)):
test_leaf = test_leaves[test_leaf_idx] #array of leaf values
true_label = yTe[test_leaf_idx]
dist_predicted_labels = [
] # List of (distance, predicated_label) pairs
for training_leaf_idx in range(len(training_leaves)):
training_leaf = training_leaves[
training_leaf_idx] #array of leaf values
predicted_label = yTr[training_leaf_idx]
distance = numpy.sum(training_leaf != test_leaf) / float(
training_leaf.size)
if distance == 1.0:
continue
dist_predicted_labels.append(
(distance,
predicted_label)) # tuple(distance, predicted_label)
closest_distances_labels = sorted(
dist_predicted_labels
)[:knn] #array of tuples (distance, predicted_label)
# vote function
labels = [label for _, label in closest_distances_labels]
if len(set(labels)) == 1:
classified_label = labels[0]
else:
classified_label = config.binaryLabels[1] #webpageNonMon
debugInfo.append([true_label,
classified_label]) # for debug purposes
if true_label != config.binaryLabels[
1] and true_label == classified_label:
true_positive += 1
if true_label == config.binaryLabels[
1] and true_label != classified_label:
false_positive += 1
num_unmonitored_test_instances = yTe.count(config.binaryLabels[1])
num_monitored_test_instances = len(
yTe) - num_unmonitored_test_instances
true_positive_rate = true_positive / float(
num_monitored_test_instances)
false_positive_rate = false_positive / float(
num_unmonitored_test_instances)
print("True Positive Count = %d / %d" %
(true_positive, num_monitored_test_instances))
print("False Positive Count = %d / %d" %
(false_positive, num_unmonitored_test_instances))
print("True Positive Rate: ", true_positive_rate)
print("False Positive Rate: ", false_positive_rate)
result = [true_positive_rate, false_positive_rate]
return [result, debugInfo]
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
[trainingFileOrig, testingFileOrig] = [trainingFile, testingFile]
if config.NUM_MONITORED_SITES != -1: #no need to classify as this is for generating openworld datasets. See the line above (arffWriter)
[accuracy, debugInfo] = ['NA', []]
return [accuracy, debugInfo]
if config.n_components_PCA != 0:
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
if config.n_components_LDA != 0:
[trainingFile,
testingFile] = Utils.calcLDA4([trainingFile, testingFile])
if config.n_components_QDA != 0:
[trainingFile,
testingFile] = Utils.calcQDA([trainingFile, testingFile])
if config.lasso != 0:
#[trainingFile,testingFile] = Utils.calcLasso3([trainingFile,testingFile])
#[trainingFile,testingFile] = Utils.calcLogisticRegression([trainingFile,testingFile])
Utils.calcLogisticRegression([trainingFile, testingFile])
# deep learning
if config.DEEP_LEARNING_METHOD != -1:
#[trainingFile, testingFile] = dA_2.calcAE([trainingFile, testingFile]) # one layer dA
#[trainingFile, testingFile] = dA_2.calcAE([trainingFile, testingFile]) # two layers dA
#[trainingFile, testingFile] = dA_2.calcAE([trainingFile, testingFile])
#SdA_2.calcSdA([trainingFile, testingFile])
if config.DEEP_LEARNING_METHOD == 1:
[trainingFile, testingFile
] = logistic_sgd_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 2:
[trainingFile,
testingFile] = dA_2.runDL([trainingFile, testingFile])
[trainingFile,
testingFile] = dA_2.runDL([trainingFile, testingFile])
#[trainingFile, testingFile] = dA_2.runDL([trainingFile, testingFile])
#[trainingFile, testingFile] = dA_2.runDL([trainingFile, testingFile])
#[trainingFile, testingFile] = dA_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 3:
# DL classifier
return mlp_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 4:
return SdA_2.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 5:
return mlp_3.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 6:
return SdA_3.runDL([trainingFile, testingFile])
elif config.DEEP_LEARNING_METHOD == 7:
return LeNetConvPoolLayer_2.runDL([trainingFile, testingFile])
#Utils.plotDensity([trainingFile,testingFile])
#Utils.plot([trainingFile,testingFile])
if config.OC_SVM == 0: # multi-class svm
if config.CROSS_VALIDATION == 0:
#print 'WARNING: NB classifier with Bi-Di. ###########///////////XXXXXX???????? '
#return wekaAPI.execute(trainingFile, testingFile, "weka.classifiers.bayes.NaiveBayes", ['-K'])
return wekaAPI.execute(
trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
#'-S','2', # one-class svm
'-B'
]) # confidence
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
'-B'
]) # confidence
else: # one-class svm
if config.CROSS_VALIDATION == 0:
print str(config.SVM_KERNEL)
print str(config.OC_SVM_Nu)
return wekaAPI.executeOneClassSVM(
trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-K',
str(config.SVM_KERNEL),
#'-K','2', # RBF kernel
#'-G','0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
#'-C','131072', # Cost
#'-N','0.001', # nu
'-N',
str(config.OC_SVM_Nu), # nu
'-S',
'2'
]) #, # one-class svm
#'-B'] ) # confidence
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
'-B'
]) # confidence
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
if config.CROSS_VALIDATION == 0:
trainList = wekaAPI.readFile(trainingFile)
testList = wekaAPI.readFile(testingFile)
trainInstancesList = []
testInstancesList = []
classes = ""
yTrain = []
yTest = []
for line in trainList:
if line[0] == '@':
if line.lower().startswith("@attribute class"):
classes = line.split(" ")[2]
else:
# instancesList.append(float(line.split(",")[:-1]))
trainInstancesList.append(
[float(i) for i in line.split(",")[:-1]])
yTrain.append(line.split(",")[-1])
for line in testList:
if line[0] != '@':
testInstancesList.append(
[float(i) for i in line.split(",")[:-1]])
yTest.append(line.split(",")[-1])
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
fileList = wekaAPI.readFile(file)
fileInstancesList = []
y = []
for line in fileList:
if line[0] == '@':
if line.lower().startswith("@attribute class"):
classes = line.split(" ")[2]
else:
fileInstancesList.append(
[float(i) for i in line.split(",")[:-1]])
y.append(line.split(",")[-1])
trainInstancesList, testInstancesList, yTrain, yTest = train_test_split(
fileInstancesList, y, test_size=6000, random_state=42)
XTr = numpy.array(trainInstancesList)
yTr = numpy.array(yTrain)
XTe = numpy.array(testInstancesList)
yTe = numpy.array(yTest)
# print ("Scaling data...")
# scaler = StandardScaler()
# XTr = scaler.fit_transform(XTr)
# XTe = scaler.fit_transform(XTe)
if config.NUM_MONITORED_SITES == -1 and config.NUM_NON_MONITORED_SITES == -1:
print "Closed-world"
else:
print "Open-world"
print("Classification...")
classifier = kNN()
classifier.fit(XTr, yTr)
prediction = classifier.predict(XTe)
totalPredictions = 0
totalCorrectPredictions = 0
debugInfo = []
for i in range(0, len(yTe)):
actualClass = yTe[i]
predictedClass = prediction[i]
probEstimate = 'NA'
# debugInfo.append([actualClass,predictedClass])
debugInfo.append([actualClass, predictedClass, probEstimate])
totalPredictions += 1.0
if actualClass == predictedClass:
totalCorrectPredictions += 1.0
accuracy = totalCorrectPredictions / totalPredictions * 100.0
print("Accuracy = ", accuracy)
positive = [] # monitored
negative = [] # non-monitored
positive.append(config.binaryLabels[0]) # 'webpageMon'
negative.append(config.binaryLabels[1]) # 'webpageNonMon'
tp = 0
tn = 0
fp = 0
fn = 0
for entry in debugInfo:
if entry[0] in positive: # actual is positive
if entry[1] in positive: # predicted is positive too
tp += 1
else: # predicted is negative
fn += 1
elif entry[0] in negative: # actual is negative
if entry[1] in positive: # predicted is positive
fp += 1
else: # predicted is negative too
tn += 1
tpr = str("%.4f" % (float(tp) / float(tp + fn)))
fpr = str("%.4f" % (float(fp) / float(fp + tn)))
Acc = str("%.4f" % (float(tp + tn) / float(tp + tn + fp + fn)))
F1 = str("%.4f" % (float(2 * tp) / float((2 * tp) + (fn) + (fp))))
F2 = str("%.4f" % (float(5 * tp) / float((5 * tp) + (4 * fn) +
(fp)))) # beta = 2
print "TPR, FPR, ACC, tp, tn, fp, fn, F1, F2"
print tpr, fpr, Acc, tp, tn, fp, fn, F1, F2
return [accuracy, debugInfo]
def classify(runID, trainingSet, testingSet):
[trainingFile,
testingFile] = arffWriter.writeArffFiles(runID, trainingSet,
testingSet)
if config.NUM_MONITORED_SITES != -1: #no need to classify as this is for generating openworld datasets. See the line above (arffWriter)
[accuracy, debugInfo] = ['NA', []]
return [accuracy, debugInfo]
if config.n_components_PCA != 0:
[trainingFile,
testingFile] = Utils.calcPCA2([trainingFile, testingFile])
if config.n_components_LDA != 0:
[trainingFile,
testingFile] = Utils.calcLDA4([trainingFile, testingFile])
if config.n_components_QDA != 0:
[trainingFile,
testingFile] = Utils.calcQDA([trainingFile, testingFile])
if config.lasso != 0:
#[trainingFile,testingFile] = Utils.calcLasso3([trainingFile,testingFile])
#[trainingFile,testingFile] = Utils.calcLogisticRegression([trainingFile,testingFile])
Utils.calcLogisticRegression([trainingFile, testingFile])
#Utils.plotDensity([trainingFile,testingFile])
#Utils.plot([trainingFile,testingFile])
if config.NUM_FEATURES_RF != 0:
[trainingFile,
testingFile] = Utils.calcTreeBaseRF([trainingFile, testingFile],
config.NUM_FEATURES_RF)
if config.OC_SVM == 0: # multi-class svm
if config.CROSS_VALIDATION == 0:
return wekaAPI.execute(
trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
'-B'
]) # confidence
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
'-B'
]) # confidence
else: # one-class svm
if config.CROSS_VALIDATION == 0:
print str(config.SVM_KERNEL)
print str(config.OC_SVM_Nu)
return wekaAPI.executeOneClassSVM(
trainingFile,
testingFile,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-K',
str(config.SVM_KERNEL),
#'-K','0', # kernel
#'-G','0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
#'-C','131072', # Cost
#'-N','0.01', # nu
'-N',
str(config.OC_SVM_Nu), # nu
'-S',
'2'
]) #, # one-class svm
#'-B'] ) # confidence
else:
file = Utils.joinTrainingTestingFiles(
trainingFile, testingFile) # join and shuffle
return wekaAPI.executeCrossValidation(
file,
"weka.Run weka.classifiers.functions.LibSVM",
[
'-x',
str(config.CROSS_VALIDATION), # number of folds
'-K',
'2', # RBF kernel
'-G',
'0.0000019073486328125', # Gamma
##May20 '-Z', # normalization 18 May 2015
'-C',
'131072', # Cost
'-B'
]) # confidence
