def mlecsvm(unlabel, clfs, true, x, y, test):
printOn.blockPrint()
noisy_labels, confusion_matrixs, count_vi, answer = wrapperDS.run(
unlabel, clfs, true)
printOn.enablePrint()
df_noise_x, df_noise_y, noiseLabel = shuffle.run(unlabel, noisy_labels, x,
y)
ground = []
X_test = []
for point in test:
ground.append(int(point[0]))
X_test.append(point[1:])
clf = CSVM.SVM(C=1000.1)
X_train, y_train = np.asarray(df_noise_x), np.asarray(df_noise_y)
printOn.blockPrint()
clf.fit(X_train, y_train)
X_test = np.asarray(X_test)
y_predict = clf.predict(X_test)
predict = y_predict.tolist()
printOn.enablePrint()
count = 0
for index in range(len(ground)):
est = predict[index]
truth = ground[index]
if est != truth:
count += 1
return count / len(ground)
def mlenoiseboost_viz(percentage):
unlabel, clfs, true, x, y, test = cancer.preprocess(percentage)
printOn.blockPrint()
one = 0
zero = 0
for label in y:
if int(label) == 1:
one += 1
else:
zero += 1
one /= len(y)
zero /= len(y)
noisy_labels, confusion_matrixs, count_vi, answer = wrapperDS.run(
unlabel, clfs, true)
printOn.enablePrint()
df_noise_x, df_noise_y, noiseLabel = shuffle.run(unlabel, noisy_labels, x,
y)
# get boosting ready
ones, agree = boosting.probab(answer, noisy_labels, 0)
err_pos, err_neg = boosting.calcError(confusion_matrixs, ones, agree, one,
zero)
df_noise_x = np.asarray(df_noise_x)
df_noise_y = np.asarray(df_noise_y)
# run adaboosting
clfs1 = boosting.adaboost_clf(df_noise_y, df_noise_x, 20, err_pos, err_neg,
noiseLabel)
return df_noise_x, df_noise_y, clfs1
def mlenoiseboost(unlabel, clfs, true, x, y, test):
printOn.blockPrint()
one = 0
zero = 0
for label in y:
if int(label) == 1:
one += 1
else:
zero += 1
one /= len(y)
zero /= len(y)
noisy_labels, confusion_matrixs, count_vi, answer = wrapperDS.run(
unlabel, clfs, true)
printOn.enablePrint()
df_noise_x, df_noise_y, noiseLabel = shuffle.run(unlabel, noisy_labels, x,
y)
# get boosting ready
ones, agree = boosting.probab(answer, noisy_labels, 0)
err_pos, err_neg = boosting.calcError(confusion_matrixs, ones, agree, one,
zero)
lenU = len(unlabel)
errt = (err_neg + err_pos) * 100
val = (count_vi / lenU) * 100
# run adaboosting
clfs1 = boosting.adaboost_clf(df_noise_y, df_noise_x, 20, err_pos, err_neg,
noiseLabel)
# caclulate the error rate
err1 = errorTest.test(clfs1, test, 1)
return err1
def preprocess(percentage, basicNN=False):
printOn.blockPrint()
if basicNN == True:
test, unlabel, label, true, x, y, x_true, y_true, x_test, y_test = read.read(
file='diabetes.csv',
drop=None,
retNum=1,
chopNum=1,
unlabel_percentage=percentage,
ytrain=True)
else:
test, unlabel, label, true, x, y, x_true, y_true = read.read(
file='diabetes.csv',
drop=None,
retNum=1,
chopNum=1,
unlabel_percentage=percentage)
clfs = classifiers.ensemble(x, y)
printOn.enablePrint()
for point in test:
point.insert(0, point.pop())
if basicNN == True:
return unlabel, clfs, true, x, y, test, y_test, x_test
else:
return unlabel, clfs, true, x, y, test
def mleadaboost(unlabel, clfs, true, x, y, test):
printOn.blockPrint()
noisy_labels, confusion_matrixs, count_vi, answer = wrapperDS.run(
unlabel, clfs, true)
printOn.enablePrint()
df_noise_x, df_noise_y, noiseLabel = shuffle.run(unlabel, noisy_labels, x,
y)
bdt = AdaBoostClassifier(DecisionTreeClassifier(max_depth=1),
algorithm="SAMME",
n_estimators=20)
bdt.fit(df_noise_x, df_noise_y)
err1 = errorTest.test(bdt, test, 2)
return err1
def preprocess(percentage, basicNN=False):
printOn.blockPrint()
if basicNN == True:
test, unlabel, label, true, x, y, x_true, y_true, x_test, y_test = read.read(
file='data.csv',
drop=['id'],
retNum=1,
chopNum=0,
unlabel_percentage=percentage,
ytrain=True)
else:
test, unlabel, label, true, x, y, x_true, y_true = read.read(
file='data.csv',
drop=['id'],
retNum=1,
chopNum=0,
unlabel_percentage=percentage)
clfs = classifiers.ensemble(x, y)
printOn.enablePrint()
if basicNN == True:
return unlabel, clfs, true, x, y, test, y_test, x_test
else:
return unlabel, clfs, true, x, y, test