程序簡介
程序調用tensorflow.keras搭建了一個簡單長短記憶型網絡(LSTM)，以上證指數為例，對數據進行標准化處理，輸入5天的'收盤價', '最高價', '最低價','開盤價'，輸出1天的'收盤價'，利用訓練集訓練網絡后，輸出測試集的MAE
長短記憶型網絡(LSTM)：是一種改進之后的循環神經網絡，可以解決RNN無法處理長距離的依賴的問題。
程序/數據集下載
代碼分析
導入模塊、路徑
# -*- 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
#用來正常顯示中文標簽
plt.rcParams['font.sans-serif']=['SimHei']
#用來正常顯示負號
plt.rcParams['axes.unicode_minus']=False
#路徑目錄
baseDir = ''#當前目錄
staticDir = os.path.join(baseDir,'Static')#靜態文件目錄
resultDir = os.path.join(baseDir,'Result')#結果文件目錄
讀取數據，查看5行
#讀取數據
data = pd.read_csv(staticDir+'/000001.csv',encoding='gbk').iloc[-100:,:]
data = data.set_index(['日期'])
data.head()
| 股票代碼 | 名稱 | 收盤價 | 最高價 | 最低價 | 開盤價 | 前收盤 | 漲跌額 | 漲跌幅 | 成交量 | 成交金額 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 日期 | |||||||||||
| 2019/9/16 | '000001 | 上證指數 | 3030.7544 | 3042.9284 | 3020.0495 | 3041.9220 | 3031.2351 | -0.4807 | -0.0159 | 221878959 | 2.37E+11 |
| 2019/9/17 | '000001 | 上證指數 | 2978.1178 | 3023.7109 | 2970.5704 | 3023.7109 | 3030.7544 | -52.6366 | -1.7367 | 223338061 | 2.38E+11 |
| 2019/9/18 | '000001 | 上證指數 | 2985.6586 | 2996.4022 | 2982.4003 | 2984.0837 | 2978.1178 | 7.5408 | 0.2532 | 168046699 | 2.00E+11 |
| 2019/9/19 | '000001 | 上證指數 | 2999.2789 | 2999.2789 | 2975.3978 | 2992.9222 | 2985.6586 | 13.6203 | 0.4562 | 162690615 | 1.93E+11 |
| 2019/9/20 | '000001 | 上證指數 | 3006.4467 | 3011.3400 | 2996.1929 | 3004.8142 | 2999.2789 | 7.1678 | 0.239 | 182145302 | 2.18E+11 |
對輸入輸出進行標准化，查看5行
#標准化數據集
outputCol = ['收盤價']#輸出列
inputCol = ['收盤價', '最高價','最低價','開盤價']#輸入列
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 ]])
將數據按時間步進行整理，時間步這里設置為5天，輸入為1天
#按時間步組成輸入輸出集
timeStep = 5#輸入天數
outStep = 1#輸出天數
xAll = list()
yAll = list()
#按時間步整理數據 輸入數據尺寸是(timeStep,5) 輸出尺寸是(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('輸入集尺寸',xAll.shape)
print('輸出集尺寸',yAll.shape)
輸入集尺寸 (95, 5, 4)
輸出集尺寸 (95, 1)
數據集分割為訓練集和測試集
#分成測試集，訓練集
testRate = 0.2#測試比例
splitIndex = int(xAll.shape[0]*(1-testRate))
xTrain = xAll[:splitIndex]
xTest = xAll[splitIndex:]
yTrain = yAll[:splitIndex]
yTest = yAll[splitIndex:]
搭建一個簡單的LSTM網絡，結構下文會打印出來
def buildLSTM(timeStep,inputColNum,outStep,learnRate=1e-4):
'''
搭建LSTM網絡，激活函數為tanh
timeStep：輸入時間步
inputColNum：輸入列數
outStep：輸出時間步
learnRate：學習率
'''
#輸入層
inputLayer = Input(shape=(timeStep,inputColNum))
#中間層
middle = LSTM(100,activation='tanh')(inputLayer)
middle = Dense(100,activation='tanh')(middle)
#輸出層 全連接
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
_________________________________________________________________
利用訓練集對網絡進行訓練
#訓練網絡
epochs = 1000#迭代次數
batchSize = 500#批處理量
lstm.fit(xTrain,yTrain,epochs=epochs,verbose=0,batch_size=batchSize)
對測試集進行預測，保存預測結果，查看5行
#預測 測試集對比
yPredict = lstm.predict(xTest)
yPredict = yScaler.inverse_transform(yPredict)[:,0]
yTest = yScaler.inverse_transform(yTest)[:,0]
result = {'觀測值':yTest,'預測值':yPredict}
result = pd.DataFrame(result)
result.index = data.index[timeStep+xTrain.shape[0]:result.shape[0]+timeStep+xTrain.shape[0]]
result.to_excel(resultDir+'/預測結果.xlsx')
result.head()
| 觀測值 | 預測值 | |
|---|---|---|
| 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 |
計算測試集MAE，進行可視化
mae = MAE(result['觀測值'],result['預測值'])
print('模型測試集MAE',mae)
#可視化
fig,ax = plt.subplots(1,1)
ax.plot(result.index,result['預測值'],label='預測值')
ax.plot(result.index,result['觀測值'],label='觀測值')
ax.set_title('LSTM預測效果，MAE：%2f'%mae)
ax.legend()
ax.xaxis.set_major_locator(ticker.MultipleLocator(5))
fig.savefig(resultDir+'/預測折線圖.png',dpi=500)
模型測試集MAE 37.06394592927633
