长短时记忆神经网络python代码_python调用tensorflow.keras搭建长短记忆型网络(LSTM)——以预测股票收盘价为例...

程序简介

程序调用tensorflow.keras搭建了一个简单长短记忆型网络(LSTM)，以上证指数为例，对数据进行标准化处理，输入5天的'收盘价', '最高价','最低价','开盘价'，输出1天的'收盘价'，利用训练集训练网络后，输出测试集的MAE

长短记忆型网络(LSTM)：是一种改进之后的循环神经网络，可以解决RNN无法处理长距离的依赖的问题。

程序/数据集下载

代码分析

导入模块、路径

# -*- coding: utf-8 -*-

from sklearn.preprocessing import StandardScaler

from tensorflow.keras.layers import Input,Dense,LSTM,GRU,BatchNormalization

from tensorflow.keras.layers import PReLU

from tensorflow.keras.models import Model

from tensorflow.keras.optimizers import Adam

from sklearn.metrics import mean_absolute_error as MAE

import matplotlib.pyplot as plt

import matplotlib.ticker as ticker

import pandas as pd

import numpy as np

import os

#用来正常显示中文标签

plt.rcParams['font.sans-serif']=['SimHei']

#用来正常显示负号

plt.rcParams['axes.unicode_minus']=False

#路径目录

baseDir = ''#当前目录

staticDir = os.path.join(baseDir,'Static')#静态文件目录

resultDir = os.path.join(baseDir,'Result')#结果文件目录

读取数据，查看5行

#读取数据

data = pd.read_csv(staticDir+'/000001.csv',encoding='gbk').iloc[-100:,:]

data = data.set_index(['日期'])

data.head()

股票代码

名称

收盘价

最高价

最低价

开盘价

前收盘

涨跌额

涨跌幅

成交量

成交金额

日期

2019/9/16

'000001

上证指数

3030.7544

3042.9284

3020.0495

3041.9220

3031.2351

-0.4807

-0.0159

221878959

2.37E+11

2019/9/17

'000001

上证指数

2978.1178

3023.7109

2970.5704

3023.7109

3030.7544

-52.6366

-1.7367

223338061

2.38E+11

2019/9/18

'000001

上证指数

2985.6586

2996.4022

2982.4003

2984.0837

2978.1178

7.5408

0.2532

168046699

2.00E+11

2019/9/19

'000001

上证指数

2999.2789

2999.2789

2975.3978

2992.9222

2985.6586

13.6203

0.4562

162690615

1.93E+11

2019/9/20

'000001

上证指数

3006.4467

3011.3400

2996.1929

3004.8142

2999.2789

7.1678

0.239

182145302

2.18E+11

对输入输出进行标准化，查看5行

#标准化数据集

outputCol = ['收盘价']#输出列

inputCol = ['收盘价', '最高价','最低价','开盘价']#输入列

X = data[inputCol]

Y = data[outputCol]

xScaler = StandardScaler()

yScaler = StandardScaler()

X = xScaler.fit_transform(X)

Y = yScaler.fit_transform(Y)

X[:5,:]

array([[0.94704786, 0.91606531, 0.98497021, 1.04253169],

[0.21175964, 0.65151178, 0.33108448, 0.80913257],

[0.31709816, 0.2755725 , 0.48742125, 0.30125807],

[0.50736208, 0.31517397, 0.39488046, 0.41453503],

[0.60749011, 0.48121048, 0.66969587, 0.5669466 ]])

将数据按时间步进行整理，时间步这里设置为5天，输入为1天

#按时间步组成输入输出集

timeStep = 5#输入天数

outStep = 1#输出天数

xAll = list()

yAll = list()

#按时间步整理数据 输入数据尺寸是(timeStep,5) 输出尺寸是(outSize)

for row in range(data.shape[0]-timeStep-outStep+1):

x = X[row:row+timeStep]

y = Y[row+timeStep:row+timeStep+outStep]

xAll.append(x)

yAll.append(y)

xAll = np.array(xAll).reshape(-1,timeStep,len(inputCol))

yAll = np.array(yAll).reshape(-1,outStep)

print('输入集尺寸',xAll.shape)

print('输出集尺寸',yAll.shape)

输入集尺寸 (95, 5, 4)

输出集尺寸 (95, 1)

数据集分割为训练集和测试集

#分成测试集，训练集

testRate = 0.2#测试比例

splitIndex = int(xAll.shape[0]*(1-testRate))

xTrain = xAll[:splitIndex]

xTest = xAll[splitIndex:]

yTrain = yAll[:splitIndex]

yTest = yAll[splitIndex:]

搭建一个简单的LSTM网络，结构下文会打印出来

def buildLSTM(timeStep,inputColNum,outStep,learnRate=1e-4):

'''

搭建LSTM网络，激活函数为tanh

timeStep：输入时间步

inputColNum：输入列数

outStep：输出时间步

learnRate：学习率

'''

#输入层

inputLayer = Input(shape=(timeStep,inputColNum))

#中间层

middle = LSTM(100,activation='tanh')(inputLayer)

middle = Dense(100,activation='tanh')(middle)

#输出层 全连接

outputLayer = Dense(outStep)(middle)

#建模

model = Model(inputs=inputLayer,outputs=outputLayer)

optimizer = Adam(lr=learnRate)

model.compile(optimizer=optimizer,loss='mse')

model.summary()

return model

#搭建LSTM

lstm = buildLSTM(timeStep=timeStep,inputColNum=len(inputCol),outStep=outStep,learnRate=1e-4)

_________________________________________________________________

Layer (type) Output Shape Param #

=================================================================

input_1 (InputLayer) (None, 5, 4) 0

_________________________________________________________________

lstm (LSTM) (None, 100) 42000

_________________________________________________________________

dense (Dense) (None, 100) 10100

_________________________________________________________________

dense_1 (Dense) (None, 1) 101

=================================================================

Total params: 52,201

Trainable params: 52,201

Non-trainable params: 0

_________________________________________________________________

利用训练集对网络进行训练

#训练网络

epochs = 1000#迭代次数

batchSize = 500#批处理量

lstm.fit(xTrain,yTrain,epochs=epochs,verbose=0,batch_size=batchSize)

对测试集进行预测，保存预测结果，查看5行

#预测 测试集对比

yPredict = lstm.predict(xTest)

yPredict = yScaler.inverse_transform(yPredict)[:,0]

yTest = yScaler.inverse_transform(yTest)[:,0]

result = {'观测值':yTest,'预测值':yPredict}

result = pd.DataFrame(result)

result.index = data.index[timeStep+xTrain.shape[0]:result.shape[0]+timeStep+xTrain.shape[0]]

result.to_excel(resultDir+'/预测结果.xlsx')

result.head()

观测值

预测值

日期

2020/1/15

3090.0379

3119.753662

2020/1/16

3074.0814

3103.595947

2020/1/17

3075.4955

3085.278809

2020/1/20

3095.7873

3079.762451

2020/1/21

3052.1419

3094.907471

计算测试集MAE，进行可视化

mae = MAE(result['观测值'],result['预测值'])

print('模型测试集MAE',mae)

#可视化

fig,ax = plt.subplots(1,1)

ax.plot(result.index,result['预测值'],label='预测值')

ax.plot(result.index,result['观测值'],label='观测值')

ax.set_title('LSTM预测效果，MAE：%2f'%mae)

ax.legend()

ax.xaxis.set_major_locator(ticker.MultipleLocator(5))

fig.savefig(resultDir+'/预测折线图.png',dpi=500)

模型测试集MAE 37.06394592927633