1 '''神经网络测试'''
2 import pandas as pd
3 from keras.models import Sequential
4 from keras.layers.core import Dense, Activation
5 import numpy as np
6
7 # 参数初始化
8 inputfile = 'C:/Users/linji/Desktop/bankloan.xls'
9 data = pd.read_excel(inputfile)
10 x_test = data.iloc[:,:8].values
11 y_test = data.iloc[:,8].values
12
13 model = Sequential() # 建立模型
14 model.add(Dense(input_dim = 8, units = 8))
15 model.add(Activation('relu')) # 用relu函数作为激活函数，能够大幅提供准确度
16 model.add(Dense(input_dim = 8, units = 1))
17 model.add(Activation('sigmoid')) # 由于是0-1输出，用sigmoid函数作为激活函数
18
19 model.compile(loss = 'mean_squared_error', optimizer = 'adam')
20 # 编译模型。由于我们做的是二元分类，所以我们指定损失函数为binary_crossentropy，以及模式为binary
21 # 另外常见的损失函数还有mean_squared_error、categorical_crossentropy等，请阅读帮助文件。
22 # 求解方法我们指定用adam，还有sgd、rmsprop等可选
23
24 model.fit(x_test, y_test, epochs = 1000, batch_size = 10)
25
26 predict_x=model.predict(x_test)
27 classes_x=np.argmax(predict_x,axis=1)
28 yp = classes_x.reshape(len(y_test))
29
30 def cm_plot(y, yp):
31
32 from sklearn.metrics import confusion_matrix #µ¼Èë»ìÏý¾ØÕóº¯Êý
33
34 cm = confusion_matrix(y, yp) #»ìÏý¾ØÕó
35
36 import matplotlib.pyplot as plt #µ¼Èë×÷ͼ¿â
37 plt.matshow(cm, cmap=plt.cm.Greens) #»»ìÏý¾ØÕóͼ£¬ÅäÉ«·ç¸ñʹÓÃcm.Greens£¬¸ü¶à·ç¸ñÇë²Î¿¼¹ÙÍø¡£
38 plt.colorbar() #ÑÕÉ«±êÇ©
39
40 for x in range(len(cm)): #Êý¾Ý±êÇ©
41 for y in range(len(cm)):
42 plt.annotate(cm[x,y], xy=(x, y), horizontalalignment='center', verticalalignment='center')
43
44 plt.ylabel('True label') #×ø±êÖá±êÇ©
45 plt.xlabel('Predicted label') #×ø±êÖá±êÇ©
46 return plt
47
48 cm_plot(y_test,yp).show()# 显示混淆矩阵可视化结果
49
50 score = model.evaluate(x_test,y_test,batch_size=128) # 模型评估
51 print(score)
1 # -*- coding: utf-8 -*-
2
3 # 代码5-2
4
5 import pandas as pd
6 # 参数初始化
7 filename = 'C:/Users/linji/Desktop/bankloan.xls'
8 data = pd.read_excel(filename) # 导入数据
9
10 # 数据是类别标签，要将它转换为数据
11 # 用1来表示“好”“是”“高”这三个属性，用-1来表示“坏”“否”“低”
12
13 x = data.iloc[:,:8].astype(int)
14 y = data.iloc[:,8].astype(int)
15
16
17 from sklearn.tree import DecisionTreeClassifier as DTC
18 dtc = DTC(criterion='entropy') # 建立决策树模型，基于信息熵
19 dtc.fit(x, y) # 训练模型
20
21 # 导入相关函数，可视化决策树。
22 # 导出的结果是一个dot文件，需要安装Graphviz才能将它转换为pdf或png等格式。
23 from sklearn.tree import export_graphviz
24 x = pd.DataFrame(x)
25
26 """
27 string1 = '''
28 edge [fontname="NSimSun"];
29 node [ fontname="NSimSun" size="15,15"];
30 {
31 '''
32 string2 = '}'
33 """
34
35 with open("C:/Users/linji/Desktop/tree.dot", 'w') as f:
36 export_graphviz(dtc, feature_names = x.columns, out_file = f)
37 f.close()
38
39
40 from IPython.display import Image
41 from sklearn import tree
42 import pydotplus
43
44 dot_data = tree.export_graphviz(dtc, out_file=None, #regr_1 是对应分类器
45 feature_names=data.columns[:8], #对应特征的名字
46 class_names=data.columns[8], #对应类别的名字
47 filled=True, rounded=True,
48 special_characters=True)
49
50 graph = pydotplus.graph_from_dot_data(dot_data)
51 graph.write_png('C:/Users/linji/Desktop/example.png') #保存图像
52 Image(graph.create_png())
