CNN+LSTM多通道特征组合模型_cnn_lstm多头模型

1、摘要

本文主要讲解：CNN+LSTM多通道特征组合模型

2、包版本

pip install tensorflow-gpu==2.3.0
pip install keras==2.4.3
1
2

3、相关技术

多通道CNN模型，应用于文本分类任务

4、完整代码和步骤

主运行程序入口

# coding:utf-8
import os

import numpy as np
import tensorflow as tf
from keras import Input, Model
from keras.initializers import he_normal
from keras.layers import Dense, Conv1D, LeakyReLU, Concatenate
from keras.optimizers import Adam
from sklearn.metrics import mean_absolute_error
from tensorflow.python.keras.layers import CuDNNLSTM

from my_utils.read_write import pdReadCsv

os.chdir(os.getcwd())
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
physical_devices = tf.config.experimental.list_physical_devices('GPU')
assert len(physical_devices) > 0, "Not enough GPU hardware devices available"
tf.config.experimental.set_memory_growth(physical_devices[0], True)

# 定义多通道特征组合模型
def build_multi_cr_lstm_model(ts, fea_dim):
    # 定义输入
    inputs = Input(shape=(ts, fea_dim))

    # filters:卷积核的数目(即输出的维度)  kernel_size:卷积核的空域或时域窗长度
    # strides:为卷积的步长  kernel_initializer 权值矩阵的初始化器
    cnn_left_out1 = Conv1D(filters=50, kernel_size=6, strides=3, kernel_initializer=he_normal(seed=3))(inputs)
    act_left_out1 = LeakyReLU()(cnn_left_out1)
    #  输出维度 return_sequences:若为True则返回整个序列，否则仅返回输出序列的最后一个输出
    lstm_left_out1 = CuDNNLSTM(64, return_sequences=False)(act_left_out1)

    # cnn层&lstm层2
    cnn_right_out1 = Conv1D(filters=50, kernel_size=12, strides=3, kernel_initializer=he_normal(seed=3))(inputs)
    act_right_out1 = LeakyReLU()(cnn_right_out1)

    lstm_right_out1 = CuDNNLSTM(64, return_sequences=False)(act_right_out1)

    # cnn层&lstm层3
    cnn_mid_out1 = Conv1D(filters=50, kernel_size=6, strides=2, kernel_initializer=he_normal(seed=3))(inputs)
    act_mid_out1 = LeakyReLU()(cnn_mid_out1)

    lstm_mid_out1 = CuDNNLSTM(64, return_sequences=False)(act_mid_out1)
    # Concatenate 连接三个数组
    concat_output = Concatenate(axis=1)([lstm_left_out1, lstm_mid_out1, lstm_right_out1])
    # 上层叠加新的dense层  units：代表该层的输出维度
    outputs = Dense(units=1)(concat_output)
    model_func = Model(inputs=inputs, outputs=outputs)
    model_func.compile(loss='mae', optimizer=Adam(lr=0.002, decay=0.01), metrics=['mae'])
    return model_func


# 构建训练集、预测集，训练和预测分别transform
def make_train_test_data():
    train_x, train_y, train_y_MinMax = fit_size(X_train, y_train)
    ts = 12
    test_x = train_x
    test_y = train_y

    ts_train_x = np.array([])
    ts_train_y = np.array([])

    ts_test_x = np.array([])
    ts_test_y = np.array([])

    # 构建训练数据集
    print('训练数据的原始shape：', train_x.shape)
    for i in range(train_x.shape[0]):
        if i + ts == train_x.shape[0]:
            break

        ts_train_x = np.append(ts_train_x, train_x[i: i + ts, :])

        ts_train_y = np.append(ts_train_y, train_y[i + ts])

    # 构建预测数据集
    print('预测数据的原始shape：', test_x.shape)
    for i in range(test_x.shape[0]):
        if i + ts == test_x.shape[0]:
            break

        ts_test_x = np.append(ts_test_x, test_x[i: i + ts, :])

        ts_test_y = np.append(ts_test_y, test_y[i + ts])
    train_x = ts_train_x.reshape((train_x.shape[0] - ts, ts, train_x.shape[1]))
    test_x = ts_test_x.reshape((test_x.shape[0] - ts, ts, test_x.shape[1]))
    return train_x, ts_train_y, test_x, ts_test_y, train_y_MinMax


def start():
    # 构建model
    lstm_model = build_multi_cr_lstm_model(12, data_dim)

    lstm_model.fit(train_x, train_y, epochs=128, batch_size=32)
    pred_y = lstm_model.predict(test_x)
    # 转换为真实值
    inv_pred_y = train_y_MinMax.inverse_transform(pred_y)
    inv_test_Y = train_y_MinMax.inverse_transform(test_y.reshape(len(test_y), 1))
    mae = int(mean_absolute_error(inv_test_Y, inv_pred_y))
    print('test_mae : ', mae)
    mae = str(mae)

    model_path = r'mepre\\'
    lstm_model.save(
        model_path + name[0] + '_' + name[1] + '_' + name[2] + '_' + mae + '.h5')


def fit_size(x, y):
    from sklearn import preprocessing
    x_MinMax = preprocessing.MinMaxScaler()
    y_MinMax = preprocessing.MinMaxScaler()
    x = x_MinMax.fit_transform(x)
    y = y_MinMax.fit_transform(y)
    return x, y, y_MinMax


def load_data(df_train):
    X_train = df_train.drop(['CreateTime'], axis=1)
    y_train = df_train['wap'].values.reshape(-1, 1)
    return X_train, y_train, X_train, y_train


if __name__ == '__main__':
    os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
    # 超参设置
    data_dim = 43
    src = r'E:\\'
    train_path = src + r'mee_del.csv'
    df = pdReadCsv(train_path, ',')
    df = df.replace("--", '0')
    df.fillna(0, inplace=True)
    groups = df.groupby(['Papr'])
    for name, group in groups:
        print(name)
        X_train, y_train, X_test, y_test = load_data(group)
        train_x, train_y, test_x, test_y, train_y_MinMax = make_train_test_data()
        start()

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138

5、学习链接

基于多通道CNN-BiGRU与多特征融合方法