dataY = np.array(dataY) # axis=0 means the array of the corresponding columns is spliced horizontally # axis=1 means the array of the corresponding rows is spliced vertically dataX = np.concatenate((dataX[0],dataX[1]),axis=0) dataY = np.concatenate((dataY[0],dataY[1]),axis=0) dataY = dataY.reshape((len(dataY),1)) #make sure the shape of the label data is (n_samples,1). # Devide dataset x_train, x_validation, y_train, y_validation = train_test_split(dataX, dataY, test_size=0.2, random_state=42) # In[3]: # Training model model = ensemble.AdaBoostClassifier( weak_classifier=DecisionTreeClassifier, n_weakers_limit=40) model.fit(X=x_train,y=y_train) # In[4]: # Predict y_train_pred=model.predict(X=x_train,threshold=0) y_validation_pred=model.predict(X=x_validation,threshold=0) # In[12]: # Verify the accuracy on the validation set
y = features_dataset[:, features_dataset.shape[1] - 1:]
np.save("X.npy", X)
np.save("y.npy", y)
if __name__ == "__main__":
#normalize()
#face_features = np.load("./datasets/face_features.npy")
#nonface_features = np.load("./datasets/nonface_features.npy")
#generateDateset(face_features, nonface_features)
X = np.load("X.npy")
y = np.load("y.npy")
mode = DecisionTreeClassifier(criterion='gini')
adaBoost = ensemble.AdaBoostClassifier(mode, 10)
xTrain, xValidation, yTrain, yValidation = train_test_split(
X, y, test_size=0.5, random_state=42)
m = adaBoost.n_weakers_limit
for i in range(m):
mode = adaBoost.fit(xTrain, yTrain)
xTest = adaBoost.predict(xValidation)
errorRate = 0
for j in range(xValidation.shape[0]):
if xTest[j] != yValidation[j]:
errorRate = errorRate + adaBoost.weight[j]
if errorRate > 0.5:
break
alpha = math.log((1 - errorRate) / errorRate) / 2
z = 0
for k in range(adaBoost.weight.shape[0]):
import numpy as np
import pickle
import ensemble
from sklearn.metrics import classification_report
if __name__ == "__main__":
#划分训练集与验证集
print('loading features!')
dataX = pickle.load(open('trainingX.data','rb'))
dataY = pickle.load(open('trainingY.data','rb'))
print('load success!')
print(dataX)
print(dataX.shape)
x_train = dataX[0:900]
y_train = dataY[0:900].flatten()
x_valid = dataX[900:1000]
y_valid = dataY[900:1000].flatten()
classifier = ensemble.AdaBoostClassifier('sklearn.tree.DecisionTreeClassifier',3)
classifier.fit(x_train,y_train)
y_pred = classifier.predict(x_valid,0)
print(classification_report(y_valid,y_pred))
with open('laclassifier_report1.txt','w+') as f:
f.write(classification_report(y_valid, y_pred))
f.close()
x_train = train[:, 0: -1]
y_train = train[:, -1]
x_test = test[:, 0: -1]
y_test = test[:, -1]
return x_train, y_train, x_test, y_test
#Begin to train
if __name__ == "__main__":
process.process_data()
x_train, y_train, x_test, y_test = get_data()
adaBoost = ensemble.AdaBoostClassifier(DecisionTreeClassifier, 1)
adaBoost.fit(x_train, y_train)
ytest_ = adaBoost.predict(x_test)
print(classification_report(y_test, ytest_))
adaBoost = ensemble.AdaBoostClassifier(DecisionTreeClassifier, 2)
adaBoost.fit(x_train, y_train)
ytest_ = adaBoost.predict(x_test)
print(classification_report(y_test, ytest_))
adaBoost = ensemble.AdaBoostClassifier(DecisionTreeClassifier, 5)
adaBoost.fit(x_train, y_train)
ytest_ = adaBoost.predict(x_test)
print(classification_report(y_test, ytest_))
adaBoost = ensemble.AdaBoostClassifier(DecisionTreeClassifier, 10)
for i in range(sample_num):
extract_feature(pic_type='face', feature_type=face_feature)
extract_feature(pic_type='nonface', feature_type=nonface_feature)
# 将数据集切分为训练集和验证集
temp_pos = np.ones(sample_num)
temp_neg = -temp_pos
all_attribute = np.concatenate((face_feature, nonface_feature), axis=0)
all_label = np.concatenate((temp_pos.reshape(-1, 1), temp_neg.reshape(-1, 1)),
axis=0)
X_train, X_valid, Y_train, Y_valid = train_test_split(all_attribute,
all_label,
test_size=0.2)
# 建立强分类器
clf = ensemble.AdaBoostClassifier(weak_classifier=weak_clf,
n_weakers_limit=weak_clf_num)
clf.fit(X_train, Y_train)
# 使用强分类器对验证集进行预测
y_pred = clf.predict(X_valid)
y_pred = y_pred.reshape(-1).tolist()
y_true = Y_valid.reshape(-1).tolist()
# 写入预测结果
fout = codecs.open('classifier_report.txt', 'w', 'utf-8')
target_names = ['face', 'nonface']
result = classification_report(y_true, y_pred, target_names=target_names)
fout.write(result)
fout.close()
print(result)
X_train, X_val, y_train, y_val = train_test_split(X,
y,
test_size=0.2,
random_state=214)
DTC = tree.DecisionTreeClassifier(max_depth=2)
# 下面为训练一个新的模型，基分类器的数量从2到10
# 为了加深迭代过程理解，此处决策树的深度设置为2或4
# for i in range(2, 10):
# My_AdaBoost = ensemble.AdaBoostClassifier(DTC, i)
# My_AdaBoost.fit(X_train, y_train)
# My_AdaBoost.predict(X_val)
# My_AdaBoost.is_good_enough(y_val)
# My_AdaBoost.save(My_AdaBoost, '2_val_' + str(i) + '_model')
# 下面为读取已训练好的模型并进行预测
My_AdaBoost = ensemble.AdaBoostClassifier(DTC)
f = open('Tree_depth_4_reports.txt', 'a')
acc = []
for i in range(2, 10):
My_model = My_AdaBoost.load('model/4_' + str(i) + '_model')
My_model.predict(X_train)
My_model.is_good_enough(y_train)
y_predict = My_model.predict(X_val)
accuray = accuracy_score(y_val, y_predict)
acc.append(accuray)
# print accuray
report = My_model.is_good_enough(y_val)
f.write('Classifier_Num = ' + str(i) + ':\n' + str(report) + '\n\n')
f.close()
draw_predict(acc)
neg_dir = pwd + 'datasets/original/nonface/' #反例目录
#从文件导入特征矩阵,如果文件还没生成(特征未提取),则提取特征并保存在文件中
if not os.path.exists('a.npy'):
transeToNPD(pos_dir, neg_dir)
else:
X, y = loadDataSet()
#将特征集和标记集切分,33.3%划分为测试集
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.33)
#声明决策树
DTC = tree.DecisionTreeClassifier(max_depth=6)
#声明Adaboost类,base_learner的最大个数为10
myAdaBoost = ensemble.AdaBoostClassifier(DTC, 10)
#训练AdaBoost模型
myAdaBoost.fit(X_train, y_train)
#保存AdaBoost模型
myAdaBoost.save(myAdaBoost, 'train_model')
#预测测试集,产生的标记集保存在AdaBoost类中.
myAdaBoost.predict(X_val)
validation = myAdaBoost.is_good_enough(y_val)
print(validation)
fh = open('report.txt', 'w')
fh.write(validation)
fh.close()
import numpy as np
import ensemble
from sklearn import tree
labels = np.load('labels.npy')
features = np.load('features.npy')
train_size = 800
X = features[:train_size]
y = labels[:train_size]
rng_state = np.random.get_state()
np.random.shuffle(X)
np.random.set_state(rng_state)
np.random.shuffle(y)
ada = ensemble.AdaBoostClassifier(tree.DecisionTreeClassifier, 12)
ada.fit(X, y)
#import os
#import numpy as np
#import pickle
#from PIL import Image
#import feature
#x=[]
#path_face = "C:/Users/asus/project3/datasets/original/face/"
#for img_face in os.listdir(path_face):
# x.append(feature.NPDFeature.extract(feature.NPDFeature(np.array(Image.open(path_face+img_face).resize((24, 24)).convert('L')))))
#path_nonface = "C:/Users/asus/project3/datasets/original/nonface/"
#for img_nonface in os.listdir(path_nonface):
# x.append(feature.NPDFeature.extract(feature.NPDFeature(np.array(Image.open(path_nonface+img_nonface).resize((24, 24)).convert('L')))))
#pickle.dump(np.array(x),open('datasets.pkl','wb'))
import numpy as np
import pickle
import ensemble
from sklearn.cross_validation import train_test_split
from sklearn.tree import DecisionTreeClassifier
x=pickle.load(open('datasets.pkl','rb'))
y=np.ones(1000)
y[500:999]=-1
x_train, x_test, y_train, y_test = train_test_split(x,y,test_size=0.33,random_state=0)
DTclf = DecisionTreeClassifier(max_depth=4)
ADBTclf=ensemble.AdaBoostClassifier(DTclf,20).fit(x_train,y_train)
y_predict=ensemble.AdaBoostClassifier.predict(ADBTclf,x_test,0)
print(1-np.sum(y_predict!=y_test)/np.shape(x_test)[0])