keras多输出多输出示例（keras教程一）

参考 keras官网

问题描述：通过模型对故障单按照优先级排序并制定给正确的部门。

输入：

• 票证的标题（文本输入），
• 票证的文本正文（文本输入），以及
• 用户添加的任何标签（分类输入）

输出：

• 优先级分数介于0和1之间（sigmoid 输出），以及
• 应该处理票证的部门（部门范围内的softmax输出）

 1 import keras
 2 import numpy as np
 3 
 4 num_tags = 12  # Number of unique issue tags
 5 num_words = 10000  # 预处理文本数据时获得的词汇量
 6 num_departments = 4  # Number of departments for predictions
 7 
 8 title_input = keras.Input(
 9     shape=(None,), name="title"
10 )  # Variable-length sequence of ints
11 body_input = keras.Input(shape=(None,), name="body")  # Variable-length sequence of ints
12 tags_input = keras.Input(
13     shape=(num_tags,), name="tags"
14 )  # Binary vectors of size `num_tags`
15 
16 # Embed each word in the title into a 64-dimensional vector
17 title_features = keras.layers.Embedding(num_words, 64)(title_input)
18 # Embed each word in the text into a 64-dimensional vector
19 body_features = keras.layers.Embedding(num_words, 64)(body_input)
20 
21 # Reduce sequence of embedded words in the title into a single 128-dimensional vector
22 title_features = keras.layers.LSTM(128)(title_features)
23 # Reduce sequence of embedded words in the body into a single 32-dimensional vector
24 body_features = keras.layers.LSTM(32)(body_features)
25 
26 # Merge all available features into a single large vector via concatenation
27 x = keras.layers.concatenate([title_features, body_features, tags_input])
28 
29 # Stick a logistic regression for priority prediction on top of the features
30 priority_pred = keras.layers.Dense(1, name="priority")(x)
31 # Stick a department classifier on top of the features
32 department_pred = keras.layers.Dense(num_departments, name="department")(x)
33 
34 # Instantiate an end-to-end model predicting both priority and department
35 model = keras.Model(
36     inputs=[title_input, body_input, tags_input],
37     outputs=[priority_pred, department_pred],
38 )
39 model.summary()
40 keras.utils.plot_model(model, "multi_input_and_output_model.png", show_shapes=True)
41 
42 # model.compile(
43 #     optimizer=keras.optimizers.RMSprop(1e-3),
44 #     loss={
45 #         "priority": "binary_crossentropy",
46 #         "department": "categorical_crossentropy",
47 #     },
48 #     loss_weights=[1.0, 0.2],
49 # )
50 
51 model.compile(
52     optimizer=keras.optimizers.RMSprop(1e-3),
53     loss={
54         "priority": "binary_crossentropy",
55         "department": "categorical_crossentropy",
56     },
57     loss_weights={'priority': 1., 'department': 0.2},)
58 # Dummy input data
59 title_data = np.random.randint(num_words, size=(1280, 10))
60 body_data = np.random.randint(num_words, size=(1280, 100))
61 tags_data = np.random.randint(2, size=(1280, num_tags)).astype("float32")
62 
63 # Dummy target data
64 priority_targets = np.random.random(size=(1280, 1))
65 dept_targets = np.random.randint(2, size=(1280, num_departments))
66 
67 model.fit(
68     {"title": title_data, "body": body_data, "tags": tags_data},
69     {"priority": priority_targets, "department": dept_targets},
70     epochs=2,
71     batch_size=32,
72 )
73 model.save("path_to_my_model")
74 model = keras.models.load_model("path_to_my_model")

环境：keras==2.2.4 tensorflow==1.12.0

模型结构

 

 

模型参数

__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to 
==================================================================================================
title (InputLayer) (None, None) 0 
__________________________________________________________________________________________________
body (InputLayer) (None, None) 0 
__________________________________________________________________________________________________
embedding_1 (Embedding) (None, None, 64) 640000 title[0][0] 
__________________________________________________________________________________________________
embedding_2 (Embedding) (None, None, 64) 640000 body[0][0] 
__________________________________________________________________________________________________
lstm_1 (LSTM) (None, 128) 98816 embedding_1[0][0] 
__________________________________________________________________________________________________
lstm_2 (LSTM) (None, 32) 12416 embedding_2[0][0] 
__________________________________________________________________________________________________
tags (InputLayer) (None, 12) 0 
__________________________________________________________________________________________________
concatenate_1 (Concatenate) (None, 172) 0 lstm_1[0][0] 
lstm_2[0][0] 
tags[0][0] 
__________________________________________________________________________________________________
priority (Dense) (None, 1) 173 concatenate_1[0][0] 
__________________________________________________________________________________________________
department (Dense) (None, 4) 692 concatenate_1[0][0] 
==================================================================================================
Total params: 1,392,097
Trainable params: 1,392,097

　　

参数计算方法

该模型有前馈神经网络和LSTM。参考深度学习模型参数计算。