03、基于Keras的人工神经网络解决鸢尾花分类问题的代码实现

东方耀

03、基于Keras的人工神经网络解决鸢尾花分类问题的代码实现


准确率达到了98%


神经网络分类问题的经典数据（集鸢尾花数据集）介绍，神经网络Python库Keras的介绍
使用Pandas读取鸢尾花数据集， 使用LabelEncoder对类别标签进行编码
使用Keras创建一个用于鸢尾花分类识别的神经网络
训练用于鸢尾花分类的神经网络 解读训练输出的日志 了解如何评价神经网络的性能




ValueError: Target is multiclass but average='binary'. Please choose another average setting, one of [None, 'micro', 'macro', 'weighted'].
TypeError: 'int' object is not iterable


损失函数并不使用测试数据（test/validation data）来衡量网络的性能
损失函数用来指导网络的训练过程！使得网络的参数向损失降低的方向改变


分类问题的损失函数：hinge loss (SVM里用到)   cross-entropy loss 交叉熵


One-Hot encoding 哑编码 独热编码
非概率的解释：hinge loss
概率解释：将输出转换为概率函数 softmax  probability


示例：手动计算3分类问题的accuracy、交叉熵损失、mse均方差损失


东方老师AI官网：http://www.ai111.vip
有任何问题可联系东方老师微信：dfy_88888
【微信二维码图片】

1. # -*- coding: utf-8 -*-
   
2. __author__ = u'东方耀 微信：dfy_88888'
   
3. __date__ = '2019/10/21 9:38'
   
4. __product__ = 'PyCharm'
   
5. __filename__ = '人工神经网络解决鸢尾花分类'
   
6. 
   
7. import numpy as np
   
8. import pandas as pd
   
9. from keras.models import Sequential
   
10. from keras.layers import Dense
    
11. from keras.optimizers import Adam, SGD
    
12. from keras.wrappers.scikit_learn import KerasClassifier
    
13. from keras.utils import np_utils
    
14. from sklearn.model_selection import cross_val_score
    
15. # k 折 交叉验证
    
16. from sklearn.model_selection import KFold
    
17. # 将分类的字符串 变成 数字 0 1 2 三个类别
    
18. from sklearn.preprocessing import LabelEncoder
    
19. # 保存模型为json 利用json文件预测
    
20. from keras.models import model_from_json
    
21. from sklearn import datasets
    
22. from sklearn.model_selection import train_test_split
    
23. from sklearn.linear_model import LogisticRegressionCV, LogisticRegression
    
24. from sklearn.metrics import accuracy_score, confusion_matrix, precision_score, recall_score, f1_score, roc_auc_score
    
25. from sklearn.metrics import mean_squared_error, mean_absolute_error
    
26. 
    
27. SEED = 666
    
28. np.random.seed(SEED)
    
29. # load data
    
30. # header=0(default)表示第一行为标题行   header=None时表示原始文件没有表头数据
    
31. df = pd.read_csv(filepath_or_buffer='datasets/iris.csv', header=0)
    
32. # iris = datasets.load_iris()
    
33. # print(iris.keys())
    
34. X = df.values[:, 1:5].astype(float)
    
35. y = df.values[:, 5]
    
36. 
    
37. print(X.shape, y.ndim)
    
38. print(X[:3])
    
39. print(y[:10])
    
40. encoder = LabelEncoder()
    
41. y = encoder.fit_transform(y)
    
42. print(y, y.ndim)
    
43. Y_onehot = np_utils.to_categorical(y, num_classes=3)
    
44. print(Y_onehot.shape, Y_onehot.ndim)
    
45. print(Y_onehot[:5, :])
    
46. 
    
47. # X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=SEED)
    
48. # log_reg = LogisticRegressionCV(cv=5, n_jobs=-1, verbose=0, multi_class='ovr')
    
49. # log_reg = LogisticRegression()
    
50. # log_reg.fit(X_train, y_train)
    
51. # y_test_predict = log_reg.predict(X_test)
    
52. # print(log_reg.score(X_test, y_test))
    
53. # print(accuracy_score(y_test, y_test_predict)) # 0.9473684210526315
    
54. # print(confusion_matrix(y_test, y_test_predict))
    
55. # print(precision_score(y_test, y_test_predict, average='micro'))
    
56. # print(recall_score(y_test, y_test_predict, average='micro'))
    
57. 
    
58. 
    
59. # 定义神经网络结构
    
60. def baseline_model():
    
61.     model = Sequential(name='baseline_model_dfy')
    
62.     # input_dim=4 or input_shape=(4,) 都是可以的
    
63.     model.add(Dense(units=7, activation='tanh', input_dim=4))
    
64.     model.add(Dense(units=3, activation='softmax'))
    
65.     # 分类问题的损失 交叉熵 'categorical_crossentropy' accuracy 92%  hinge loss
    
66.     # 回归问题的损失  'mean_squared_error' accuracy 71.3%
    
67.     # epochs 从20 调大到 200  accuracy 92% ---> 98%
    
68.     # 隐藏层神经元units=7 调大到 128  accuracy 98% ---> 98% 还是恢复为7
    
69.     model.compile(optimizer=SGD(learning_rate=0.1), loss='categorical_crossentropy', metrics=['accuracy'])
    
70.     return model
    
71. 
    
72. # from keras.wrappers.scikit_learn import KerasClassifier
    
73. estimator = KerasClassifier(build_fn=baseline_model, epochs=20, batch_size=10, verbose=1)
    
74. 
    
75. # evalute
    
76. kfold = KFold(n_splits=10, shuffle=True, random_state=SEED)
    
77. result = cross_val_score(estimator, X, Y_onehot, cv=kfold)
    
78. print(baseline_model().summary())
    
79. print('Accuracy of cv:', result)
    
80. print('Accuracy of CV mean:%.2f, std:%.3f' % (result.mean(), result.std()))
    
81. 
    
82. # train model
    
83. estimator.fit(X, Y_onehot)
    
84. print('model accuracy score:', estimator.score(X, Y_onehot))
    
85. 
    
86. 
    
87. # save model
    
88. model_json = estimator.model.to_json()
    
89. with open('model.json', 'w') as json_file:
    
90.     json_file.write(model_json)
    
91. estimator.model.save_weights('model_weights.h5')
    
92. print('saved保存了 model to disk! Done')
    
93. 
    
94. # laod model and use it for predict
    
95. with open('model.json', 'r') as file:
    
96.     loaded_model_json = file.read()
    
97. 
    
98. loaded_model = model_from_json(loaded_model_json)
    
99. loaded_model.load_weights('model_weights.h5')
    
100. print('loaded加载了 model from disk! Done')
     
101. 
     
102. predicted = loaded_model.predict(X)
     
103. print('predicted probability:', predicted)
     
104. predicted_label = loaded_model.predict_classes(X)
     
105. print('predicted预测的 label:\n' + str(predicted_label))
     
106. print('实际值 label:\n' + str(y))
     
107. print('预测的准确度：', accuracy_score(y, predicted_label))
     
108. print(confusion_matrix(y, predicted_label))
     
109. print('预测的精准率：', precision_score(y, predicted_label, average='micro'))
     
110. print('预测的召回率：', recall_score(y, predicted_label, average='micro'))
     
111. print('预测的F1 score:', f1_score(y, predicted_label, average='micro'))
     
112. # print('roc_auc(area under curve):', roc_auc_score(y, y_score=?))
     
113. 
     
114. 
     
115. 
     

复制代码