def runSVMedit(C, gamma):
# initialize with seed so we get same training and test sets every time
# i.e. for cross validation
random.seed(18283835) # use this for standard validation
# random.seed(8982101)
# random.seed(12820501)
# Throwaway test to classify browser vs bot
frac = 0.8
# Open the database
conn = sqlite3.connect("mydb")
conn.text_factory = str
c = conn.cursor()
# Data
trainingData = []
testData = []
trainlist = []
testlist = []
# bots
c.execute('select uaString from data where "Type" = "Robot"')
uaslist = []
for uaString in c:
uaslist.append(uaString[0])
nrobot = len(uaslist)
random.shuffle(uaslist)
for uaString in uaslist[0 : int(round(frac * nrobot, 0))]:
trainingData.append(("Robot", uaString))
trainlist.append(uaString)
for uaString in uaslist[int(round(frac * nrobot, 0)) : len(uaslist) + 1]:
testData.append(("Robot", uaString))
testlist.append(uaString)
print "Number robots selected for training: ", int(round(frac * nrobot, 0))
# browsers
c.execute('select uaString from data where "Type" = "Browser"')
uaslist = []
for uaString in c:
uaslist.append(uaString[0])
nbrowser = len(uaslist)
random.shuffle(uaslist)
for uaString in uaslist[0 : int(round(frac * nbrowser, 0))]:
trainingData.append(("Browser", uaString))
trainlist.append(uaString)
for uaString in uaslist[int(round(frac * nbrowser, 0)) : len(uaslist) + 1]:
testData.append(("Browser", uaString))
testlist.append(uaString)
print "Number browsers selected for training: ", int(round(frac * nbrowser, 0))
# just select first 5000 elements of train data and first 1000 elements of test data
# random.shuffle(trainingData)
# random.shuffle(testData)
# trainingData = trainingData[0:1000]
# testData = testData[0:200]
# Make a Decision Problem
params = SVMParams()
params.kernelName = "edit"
params.tokenized = False
params.C = C
params.gamma = gamma
decProb = DecisionProblem("Type", params)
# If the model is not already generated, generate it
if not decProb.haveModel():
decProb.generateModel(trainingData)
decProb.saveModel()
else:
decProb.loadModel()
# Predict
correct = 0.0
total = 0.0
correct0 = 0.0
correct1 = 0.0
false1 = 0.0
false0 = 0.0
ind = 0
fmiss = open("misclass.txt", "w")
for actual, ua in testData:
prediction = decProb.decide(ua)
if actual == "Robot":
if actual == prediction:
correct0 = correct0 + 1.0
correct = correct + 1.0
else:
false1 = false1 + 1.0
fmiss.write("false Browser: " + testlist[ind] + "\n")
elif actual == "Browser":
if actual == prediction:
correct1 = correct1 + 1.0
correct = correct + 1.0
else:
false0 = false0 + 1.0
fmiss.write("false Robot: " + testlist[ind] + "\n")
total = total + 1.0
ind = ind + 1
print "ACCURACY: ", correct / total
print "False X means classifier said data was X, but it was actually something else"
print "Correct Robot: ", correct0, "False Robot: ", false0
print "Correct Browser: ", correct1, "False Browser: ", false1
fname = decProb.modelPath() + ".results"
f = open(fname, "w")
f.write(str(decProb.svm.labelMap) + "\n")
s = "ACCURACY: " + str(correct / total) + "\n"
s = s + "False X means classifier said data was X, but it was actually something else\n"
s = s + "Correct 0: " + str(correct0) + " False 0: " + str(false0) + "\n"
s = s + "Correct 1: " + str(correct1) + " False 1: " + str(false1) + "\n"
f.write(s)
accuracy = correct / total
return accuracy, correct0, false0, correct1, false1
def runSVM(C,gamma):
#initialize with seed so we get same training and test sets every time
# i.e. for cross validation
random.seed(18283835)
# Throwaway test to classify browser vs bot
frac = 0.8
# Open the database
conn = sqlite3.connect("mydb")
conn.text_factory = str
c = conn.cursor()
# Data
trainingData = [];
testData = [];
#get Families
c.execute('select Tokens, Family from data where "Type" = "Browser" AND "Family" IS NOT NULL')
uaslist = []
familylist = []
for row in c:
uaslist.append([row[0],row[1]]) # stored as list of lists
nbrowser = len(uaslist)
random.shuffle(uaslist)
for uaString in uaslist[0:int(round(frac*nbrowser,0))]:
tokens = [int(s) for s in uaString[0].split(" ")]
family = uaString[1]
trainingData.append((family, tokens))
for uaString in uaslist[int(round(frac*nbrowser,0)):len(uaslist)+1]:
tokens = [int(s) for s in uaString[0].split(" ")]
family = uaString[1]
testData.append((family, tokens))
print 'Number browsers selected for training: ', int(round(frac*nbrowser,0))
# just select first 5000 elements of train data and first 1000 elements of test data
# random.shuffle(trainingData)
# random.shuffle(testData)
# trainingData = trainingData[0:1000]
# testData = testData[0:200]
# Make a Decision Problem
params = SVMParams()
params.kernelName = "edit"
params.tokenized = True
params.C = C
params.gamma = gamma
decProb = DecisionProblem("Family", params)
# If the model is not already generated, generate it
if not decProb.haveModel():
decProb.generateModel(trainingData)
decProb.saveModel()
else:
decProb.loadModel()
# Predict
# FIXME need to build in confusion matrix for these multiclass problems
correct = 0.0
total = 0.0
for actual, ua in testData:
prediction = decProb.decide(ua)
if(actual == prediction):
correct = correct + 1.0
total = total + 1.0
print "ACCURACY: ", correct / total
fname = decProb.modelPath() + ".results"
f = open(fname,'w')
f.write(str(decProb.svm.labelMap)+'\n')
s = "ACCURACY: " + str(correct/total) + '\n'
f.write(s)
accuracy = correct/total
return accuracy
def runSVM(C, seqLen, seqLambda):
#initialize with seed so we get same training and test sets every time
# i.e. for cross validation
random.seed(18283835) # use this for standard validation
# random.seed(8982101)
# random.seed(12820501)
# Throwaway test to classify browser vs bot
frac = 0.8
# Open the database
conn = sqlite3.connect("mydb")
conn.text_factory = str
c = conn.cursor()
# Data
trainingData = []
testData = []
trainlist = []
testlist = []
# bots
c.execute('select Tokens, uaString from data where "Type" = "Robot"')
uaslist = []
for uaString in c:
uaslist.append((uaString[0], uaString[1]))
nrobot = len(uaslist)
random.shuffle(uaslist)
for Tok, uaString in uaslist[0:int(round(frac * nrobot, 0))]:
tokens = [int(s) for s in Tok.split(" ")]
trainingData.append(('Robot', tokens))
trainlist.append(uaString)
for Tok, uaString in uaslist[int(round(frac * nrobot, 0)):len(uaslist) +
1]:
tokens = [int(s) for s in Tok.split(" ")]
testData.append(('Robot', tokens))
testlist.append(uaString)
print 'Number robots selected for training: ', int(round(frac * nrobot, 0))
# browsers
c.execute('select Tokens, uaString from data where "Type" = "Browser"')
uaslist = []
for uaString in c:
uaslist.append((uaString[0], uaString[1]))
nbrowser = len(uaslist)
random.shuffle(uaslist)
for Tok, uaString in uaslist[0:int(round(frac * nbrowser, 0))]:
tokens = [int(s) for s in Tok.split(" ")]
trainingData.append(('Browser', tokens))
trainlist.append(uaString)
for Tok, uaString in uaslist[int(round(frac * nbrowser, 0)):len(uaslist) +
1]:
tokens = [int(s) for s in Tok.split(" ")]
testData.append(('Browser', tokens))
testlist.append(uaString)
print 'Number browsers selected for training: ', int(
round(frac * nbrowser, 0))
# just select first 5000 elements of train data and first 1000 elements of test data
# random.shuffle(trainingData)
# random.shuffle(testData)
# trainingData = trainingData[0:1000]
# testData = testData[0:200]
# Make a Decision Problem
params = SVMParams()
params.kernelName = "subseq"
params.dataType = "tokens"
params.C = C
params.seqLen = seqLen
params.seqLambda = seqLambda
decProb = DecisionProblem("Type", params)
# If the model is not already generated, generate it
if not decProb.haveModel():
decProb.generateModel(trainingData)
decProb.saveModel()
else:
decProb.loadModel()
# Predict
correct = 0.0
total = 0.0
correct0 = 0.0
correct1 = 0.0
false1 = 0.0
false0 = 0.0
ind = 0
fmiss = open('misclass.txt', 'w')
for actual, ua in testData:
prediction = decProb.decide(ua)
if (actual == 'Robot'):
if (actual == prediction):
correct0 = correct0 + 1.0
correct = correct + 1.0
else:
false1 = false1 + 1.0
fmiss.write('false Browser: ' + testlist[ind] + '\n')
elif (actual == 'Browser'):
if (actual == prediction):
correct1 = correct1 + 1.0
correct = correct + 1.0
else:
false0 = false0 + 1.0
fmiss.write('false Robot: ' + testlist[ind] + '\n')
total = total + 1.0
ind = ind + 1
print "ACCURACY: ", correct / total
print "False X means classifier said data was X, but it was actually something else"
print "Correct Robot: ", correct0, "False Robot: ", false0
print "Correct Browser: ", correct1, "False Browser: ", false1
fname = decProb.modelPath() + ".results"
f = open(fname, 'w')
f.write(str(decProb.svm.labelMap) + '\n')
s = "ACCURACY: " + str(correct / total) + '\n'
s = s + "False X means classifier said data was X, but it was actually something else\n"
s = s + "Correct 0: " + str(correct0) + " False 0: " + str(false0) + '\n'
s = s + "Correct 1: " + str(correct1) + " False 1: " + str(false1) + '\n'
f.write(s)
accuracy = correct / total
return accuracy, correct0, false0, correct1, false1
def runSVM(C, gamma):
#initialize with seed so we get same training and test sets every time
# i.e. for cross validation
random.seed(18283835)
# Throwaway test to classify browser vs bot
frac = 0.8
# Open the database
conn = sqlite3.connect("mydb")
conn.text_factory = str
c = conn.cursor()
# Data
trainingData = []
testData = []
#get Families
c.execute(
'select Tokens, Family from data where "Type" = "Browser" AND "Family" IS NOT NULL'
)
uaslist = []
familylist = []
for row in c:
uaslist.append([row[0], row[1]]) # stored as list of lists
nbrowser = len(uaslist)
random.shuffle(uaslist)
for uaString in uaslist[0:int(round(frac * nbrowser, 0))]:
tokens = [int(s) for s in uaString[0].split(" ")]
family = uaString[1]
trainingData.append((family, tokens))
for uaString in uaslist[int(round(frac * nbrowser, 0)):len(uaslist) + 1]:
tokens = [int(s) for s in uaString[0].split(" ")]
family = uaString[1]
testData.append((family, tokens))
print 'Number browsers selected for training: ', int(
round(frac * nbrowser, 0))
# just select first 5000 elements of train data and first 1000 elements of test data
# random.shuffle(trainingData)
# random.shuffle(testData)
# trainingData = trainingData[0:1000]
# testData = testData[0:200]
# Make a Decision Problem
params = SVMParams()
params.kernelName = "edit"
params.tokenized = True
params.C = C
params.gamma = gamma
decProb = DecisionProblem("Family", params)
# If the model is not already generated, generate it
if not decProb.haveModel():
decProb.generateModel(trainingData)
decProb.saveModel()
else:
decProb.loadModel()
# Predict
# FIXME need to build in confusion matrix for these multiclass problems
correct = 0.0
total = 0.0
for actual, ua in testData:
prediction = decProb.decide(ua)
if (actual == prediction):
correct = correct + 1.0
total = total + 1.0
print "ACCURACY: ", correct / total
fname = decProb.modelPath() + ".results"
f = open(fname, 'w')
f.write(str(decProb.svm.labelMap) + '\n')
s = "ACCURACY: " + str(correct / total) + '\n'
f.write(s)
accuracy = correct / total
return accuracy
