tushare實(shí)戰(zhàn)LSTM實(shí)現(xiàn)黃金價(jià)格預(yù)測(cè)

文章目錄

• tushare實(shí)戰(zhàn)LSTM實(shí)現(xiàn)黃金價(jià)格預(yù)測(cè)
• 拉取數(shù)據(jù)
• 數(shù)據(jù)預(yù)處理
• 訓(xùn)練模型
• 模型預(yù)測(cè)及查看效果
• • 先看整體情況
  • 選取特定的一小段查看
• 結(jié)果分析

拉取數(shù)據(jù)

老樣子，之前tushare實(shí)戰(zhàn)分析黃金與美元收益率關(guān)系的時(shí)候也是這樣,注意: pro_api中的東西是tushare的token

# 導(dǎo)入tushare import tushare as ts # 初始化pro接口 pro = ts.pro_api('xxx')# 拉取數(shù)據(jù) df = pro.fx_daily(**{"ts_code": "XAUUSD.FXCM","trade_date": "","start_date": 20160910,"end_date": 20210910,"exchange": "FXCM","limit": "","offset": "" }, fields=["ts_code","trade_date","bid_open","bid_close","bid_high","bid_low","ask_open","ask_close","ask_high","ask_low","tick_qty" ]) print(df) df.set_index('trade_date',inplace=True) df.to_csv('黃金數(shù)據(jù)2016-9-10至2021-9-10.csv')

數(shù)據(jù)預(yù)處理

LSTM模型的核心是用一個(gè)序列數(shù)據(jù)去預(yù)測(cè)未來的數(shù)據(jù)，序列數(shù)據(jù)的構(gòu)造思路: 構(gòu)造一個(gè)隊(duì)列，將每日的數(shù)據(jù)視為一個(gè)個(gè)體，當(dāng)后一個(gè)個(gè)體進(jìn)入隊(duì)列的時(shí)候，就會(huì)擠出隊(duì)首的個(gè)體，然后在每個(gè)時(shí)刻都‘拍照’記錄下隊(duì)列的情況，就可以得到一個(gè)三維數(shù)據(jù)（len, men_day, attribute）其中l(wèi)en是序列的個(gè)數(shù)，假設(shè)數(shù)據(jù)集中有100條數(shù)據(jù)，我們將5天作為一個(gè)序列，那么就會(huì)有100-5+1=96個(gè)序列，men_day就是那個(gè)5天的5，表示序列的長(zhǎng)度，attribute則是原數(shù)據(jù)的協(xié)變量（就是數(shù)據(jù)的特征）；

了解了輸入數(shù)據(jù)后，我們就要處理輸出數(shù)據(jù)，假設(shè)我們想用截止到今天的數(shù)據(jù)去預(yù)測(cè)未來五天的價(jià)格，那么構(gòu)造訓(xùn)練集的時(shí)候，就要將價(jià)格數(shù)據(jù)作為標(biāo)簽，向前shift5天；也就是在一條數(shù)據(jù)中，除了那些他原本就有的數(shù)據(jù)，還會(huì)多一個(gè)5天之后的價(jià)格。按照這個(gè)思路，寫出數(shù)據(jù)預(yù)處理函數(shù)，其中deque表示的就是那個(gè)隊(duì)列

mem_his_days就是序列的長(zhǎng)度，pre_days就是預(yù)測(cè)未來幾天的價(jià)格

def Stock_Price_LSTM_Data_Precesing(df,mem_his_days,pre_days):df.dropna(inplace=True)df.sort_index(inplace=True)df.drop(columns='ts_code',inplace=True)# 將ask_open向前移動(dòng)了pre_days天df['label'] = df['ask_open'].shift(-pre_days)from sklearn.preprocessing import StandardScalerscaler = StandardScaler()sca_X = scaler.fit_transform(df.iloc[:,:-1])from collections import dequedeq = deque(maxlen=mem_his_days)X = []for i in sca_X:deq.append(list(i))if len(deq)==mem_his_days:X.append(list(deq))X_lately = X[-pre_days:]X = X[:-pre_days]y = df['label'].values[mem_his_days-1:-pre_days]import numpy as npX = np.array(X)y = np.array(y)return X,y,X_lately

訓(xùn)練模型

終于到我們熟悉的調(diào)參環(huán)節(jié)，我調(diào)了好一會(huì)兒，最終選定的是mem_days、Lstm層數(shù)、隱藏層數(shù)、隱層神經(jīng)元個(gè)數(shù)分別是5、3、2、64，朋友們也可以自己取調(diào)一調(diào)

pre_days = 5 # mem_days = [5,10,15] # lstm_layers = [1,2] # dense_layers = [1,2,3] # units = [16,32] mem_days = [5] lstm_layers = [3] dense_layers = [2] units = [64] from tensorflow.keras.callbacks import ModelCheckpoint for the_mem_days in mem_days:for the_lstm_layers in lstm_layers:for the_dense_layers in dense_layers:for the_units in units:filepath = './models_only_problem/{val_mape:.2f}_{epoch:02d}_'+f'men_{the_mem_days}_lstm_{the_lstm_layers}_dense_{the_dense_layers}_unit_{the_units}'checkpoint = ModelCheckpoint(filepath=filepath,save_weights_only=False,monitor='val_mape',mode='min',save_best_only=True)X,y,X_lately = Stock_Price_LSTM_Data_Precesing(golden,the_mem_days,pre_days)from sklearn.model_selection import train_test_splitX_train, X_test, y_train, y_test = train_test_split(X,y,shuffle=False,test_size=0.1)from tensorflow.keras.models import Sequentialfrom tensorflow.keras.layers import LSTM, Dense, Dropoutmodel = Sequential()model.add(LSTM(the_units,input_shape=X.shape[1:],activation='relu',return_sequences=True))model.add(Dropout(0.1))for i in range(the_lstm_layers):model.add(LSTM(the_units,activation='relu',return_sequences=True))model.add(Dropout(0.1))model.add(LSTM(the_units,activation='relu'))model.add(Dropout(0.1))for i in range(the_dense_layers):model.add(Dense(the_units,activation='relu'))model.add(Dropout(0.1))model.add(Dense(1))model.compile(optimizer='adam',loss='mse',metrics=['mape'])model.fit(X_train,y_train,batch_size=32,epochs=50,validation_data=(X_test,y_test),callbacks=[checkpoint])

模型會(huì)被保存在models_only_problem文件夾下，第一個(gè)數(shù)值表示的就是預(yù)測(cè)的精度，越小越好，后面跟的就是那些超參數(shù)的值，只要找到前面數(shù)值最小的那些超參數(shù)的值，固定這些超參數(shù)即可，我的mape數(shù)值最小是1.66

模型預(yù)測(cè)及查看效果

將剛才訓(xùn)練好的最好的模型拿下來，加載模型，把需要預(yù)測(cè)的數(shù)據(jù)扔進(jìn)去

先看整體情況

from tensorflow.keras.models import load_model import matplotlib.pyplot as plt best_model = load_model('./models_only_problem/1.66_28_men_5_lstm_3_dense_2_unit_64')pre = best_model.predict(X) print(len(pre)) plt.plot(y,color='red',label='price') plt.plot(pre,color='green',label='predict') plt.show()

選取特定的一小段查看

x_time1 = y[200:300] pre_time1 = pre[200:300] plt.plot(x_time1,color='red',label='price') plt.plot(pre_time1,color='green',label='predict') plt.legend() plt.show()

結(jié)果分析

從總體上看，LSTM的擬合效果還是不錯(cuò)的，可是在一個(gè)較短的時(shí)間內(nèi)，預(yù)測(cè)數(shù)據(jù)竟然滯后于實(shí)際價(jià)格，這有可能帶來投資的偏差，一種可行的辦法就是，在數(shù)據(jù)中加入更多的屬性，數(shù)據(jù)的質(zhì)量決定了最終神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)的質(zhì)量

總結(jié)

以上是生活随笔為你收集整理的tushare实战LSTM实现黄金价格预测的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。