06、Keras如何实现多输入和多输出_代码示例

东方耀

06、Keras如何实现多输入和多输出_代码示例


Wide与Deep模型实战


子类API
功能API（函数式API）
多输入与多输出


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】

多输出：主要针对多任务学习模型（分类与回归都需要预测输出）


Multi-Input实现Demo：

1. 
   
2. import numpy as np
   
3. import matplotlib as mpl
   
4. import matplotlib.pyplot as plt
   
5. import sklearn
   
6. import pandas as pd
   
7. import os
   
8. import sys
   
9. import time
   
10. import tensorflow as tf
    
11. from tensorflow import keras
    
12. 
    
13. print(sys.version_info)
    
14. for module in mpl, np, pd, sklearn, tf, keras:
    
15.     print(module.__name__, module.__version__)
    
16. 
    
17. from sklearn.datasets import fetch_california_housing
    
18. 
    
19. housing = fetch_california_housing()
    
20. # print(housing.DESCR)
    
21. print(housing.data.shape)
    
22. print(housing.target.shape)
    
23. 
    
24. from sklearn.model_selection import train_test_split
    
25. 
    
26. x_train_all, x_test, y_train_all, y_test = train_test_split(
    
27.     housing.data, housing.target, random_state=7)
    
28. x_train, x_valid, y_train, y_valid = train_test_split(
    
29.     x_train_all, y_train_all, random_state=11)
    
30. print(x_train.shape, y_train.shape)
    
31. print(x_valid.shape, y_valid.shape)
    
32. print(x_test.shape, y_test.shape)
    
33. 
    
34. from sklearn.preprocessing import StandardScaler
    
35. 
    
36. scaler = StandardScaler()
    
37. x_train_scaled = scaler.fit_transform(x_train)
    
38. x_valid_scaled = scaler.transform(x_valid)
    
39. x_test_scaled = scaler.transform(x_test)
    
40. 
    
41. 
    
42. def create_model_function_multi_input():
    
43.     # 第一种方法：函数式api实现wide&deep模型 像使用函数一样
    
44.     print('共%d个特征，前5个特征给wide模型，后6个特征给deep模型' % x_train.shape[1])
    
45.     input_wide = keras.layers.Input(shape=(5, ))
    
46.     input_deep = keras.layers.Input(shape=(6, ))
    
47.     # 复合函数的形式
    
48.     hidden_1 = keras.layers.Dense(30, activation='relu')(input_deep)
    
49.     hidden_2 = keras.layers.Dense(30, activation='relu')(hidden_1)
    
50.     # wide model
    
51.     concat = keras.layers.concatenate(inputs=[input_wide, hidden_2])
    
52.     output = keras.layers.Dense(1)(concat)
    
53. 
    
54.     model = keras.models.Model(inputs=[input_wide, input_deep], outputs=[output], name='wide_deep_model')
    
55.     return model
    
56. 
    
57. 
    
58. model = create_model_function_multi_input()
    
59. 
    
60. model.summary()
    
61. 
    
62. model.compile(optimizer=keras.optimizers.Adam(0.001), loss='mean_squared_error')
    
63. callbacks = [
    
64.     keras.callbacks.EarlyStopping(patience=5, min_delta=1e-2)
    
65. ]
    
66. # multi_input  a list of arrays (in case the model has multiple inputs
    
67. inputs_train = [x_train_scaled[:, :5], x_train_scaled[:, -6:]]
    
68. inputs_valid = [x_valid_scaled[:, :5], x_valid_scaled[:, -6:]]
    
69. history = model.fit(inputs_train, y_train, validation_data=(inputs_valid, y_valid), epochs=50,
    
70.                     callbacks=callbacks)
    
71. print(type(history))
    
72. print(history.history)
    
73. 
    
74. 
    
75. def plot_learning_curve(history):
    
76.     # ValueError: arrays must all be same length
    
77.     # 表格型数据 要求每一列的len一致 这里即：history.history字典里每个key对应的value长度一致
    
78.     df_history = pd.DataFrame(data=history.history)
    
79.     print(df_history)
    
80.     print(df_history.index)
    
81.     print(df_history.columns)
    
82.     print(df_history.dtypes)
    
83.     df_history.plot(figsize=(8, 5))
    
84.     plt.grid(True)
    
85.     # x就是DataFrame的索引
    
86.     plt.ylim(0, 1)
    
87.     plt.show()
    
88. 
    
89. 
    
90. plot_learning_curve(history)
    
91. # Returns the loss value & metrics values for the model in test mode.
    
92. inputs_test = [x_test_scaled[:, :5], x_test_scaled[:, -6:]]
    
93. print(model.evaluate(inputs_test, y_test))
    

复制代码

Multi-Output实现Demo：

1. 
   
2. import numpy as np
   
3. import matplotlib as mpl
   
4. import matplotlib.pyplot as plt
   
5. import sklearn
   
6. import pandas as pd
   
7. import os
   
8. import sys
   
9. import time
   
10. import tensorflow as tf
    
11. from tensorflow import keras
    
12. 
    
13. print(sys.version_info)
    
14. for module in mpl, np, pd, sklearn, tf, keras:
    
15.     print(module.__name__, module.__version__)
    
16. 
    
17. from sklearn.datasets import fetch_california_housing
    
18. 
    
19. housing = fetch_california_housing()
    
20. # print(housing.DESCR)
    
21. print(housing.data.shape)
    
22. print(housing.target.shape)
    
23. 
    
24. from sklearn.model_selection import train_test_split
    
25. 
    
26. x_train_all, x_test, y_train_all, y_test = train_test_split(
    
27.     housing.data, housing.target, random_state=7)
    
28. x_train, x_valid, y_train, y_valid = train_test_split(
    
29.     x_train_all, y_train_all, random_state=11)
    
30. print(x_train.shape, y_train.shape)
    
31. print(x_valid.shape, y_valid.shape)
    
32. print(x_test.shape, y_test.shape)
    
33. 
    
34. from sklearn.preprocessing import StandardScaler
    
35. 
    
36. scaler = StandardScaler()
    
37. x_train_scaled = scaler.fit_transform(x_train)
    
38. x_valid_scaled = scaler.transform(x_valid)
    
39. x_test_scaled = scaler.transform(x_test)
    
40. 
    
41. 
    
42. def create_model_function_multi_input_output():
    
43.     # 第一种方法：函数式api实现wide&deep模型 像使用函数一样
    
44.     print('共%d个特征，前5个特征给wide模型，后6个特征给deep模型' % x_train.shape[1])
    
45.     input_wide = keras.layers.Input(shape=(5, ))
    
46.     input_deep = keras.layers.Input(shape=(6, ))
    
47.     # 复合函数的形式
    
48.     hidden_1 = keras.layers.Dense(30, activation='relu')(input_deep)
    
49.     hidden_2 = keras.layers.Dense(30, activation='relu')(hidden_1)
    
50.     # wide model
    
51.     concat = keras.layers.concatenate(inputs=[input_wide, hidden_2])
    
52.     output = keras.layers.Dense(1)(concat)
    
53.     output2 = keras.layers.Dense(1)(hidden_2)
    
54. 
    
55.     model = keras.models.Model(inputs=[input_wide, input_deep], outputs=[output, output2], name='wide_deep_model')
    
56.     return model
    
57. 
    
58. 
    
59. model = create_model_function_multi_input_output()
    
60. 
    
61. model.summary()
    
62. 
    
63. model.compile(optimizer=keras.optimizers.Adam(0.001), loss='mean_squared_error')
    
64. callbacks = [
    
65.     keras.callbacks.EarlyStopping(patience=5, min_delta=1e-2)
    
66. ]
    
67. # multi_input  a list of arrays (in case the model has multiple inputs
    
68. inputs_train = [x_train_scaled[:, :5], x_train_scaled[:, -6:]]
    
69. inputs_valid = [x_valid_scaled[:, :5], x_valid_scaled[:, -6:]]
    
70. history = model.fit(inputs_train, [y_train, y_train], validation_data=(inputs_valid, [y_valid, y_valid]), epochs=50,
    
71.                     callbacks=callbacks)
    
72. print(type(history))
    
73. print(history.history)
    
74. print(history.history.keys())
    
75. 
    
76. 
    
77. def plot_learning_curve(history):
    
78.     # ValueError: arrays must all be same length
    
79.     # 表格型数据 要求每一列的len一致 这里即：history.history字典里每个key对应的value长度一致
    
80.     df_history = pd.DataFrame(data=history.history)
    
81.     print(df_history)
    
82.     print(df_history.index)
    
83.     print(df_history.columns)
    
84.     print(df_history.dtypes)
    
85.     df_history.plot(figsize=(8, 5))
    
86.     plt.grid(True)
    
87.     # x就是DataFrame的索引
    
88.     plt.ylim(0, 1)
    
89.     plt.show()
    
90. 
    
91. 
    
92. plot_learning_curve(history)
    
93. # Returns the loss value & metrics values for the model in test mode.
    
94. inputs_test = [x_test_scaled[:, :5], x_test_scaled[:, -6:]]
    
95. print(model.evaluate(inputs_test, [y_test, y_test]))
    

复制代码

卿卿卿姐姐

学习学习，正好有一个项目要做