深度学习之搭建LSTM模型预测股价

      大家好，我是带我去滑雪！

     本期利用Google股价数据集，该数据集中GOOG_Stock_Price_Train.csv为训练集，GOOG_Stock_Price_Test.csv为测试集，里面有开盘价、最高股价、最低股价、收盘价、调整后的收盘价、成交量，2021年11月以前，可以在美国Yahoo网站下载股价历史数据，但现在对中国已经禁用了，可以去其他地方进行下载。本次使用调整后的收盘价进行预测。

目录

1、导入相关模块和数据集

2、产生训练所需的特征和标签数据

3、转换数据为（样本数，时步、特征）的张量

4、定义LSTM模型

5、使用已经训练好的LSTM模型预测股价

6、绘制真实股价与预测股价的对比图

---

1、导入相关模块和数据集

import numpy as np
import pandas as pd
from sklearn.preprocessing import MinMaxScaler
from keras.models import Sequential
from keras.layers import Dense, Dropout,LSTM,SimpleRNN,GRU

# 载入Google股价数据集 
df_train = pd.read_csv(r'E:\工作\硕士\博客\博客37-\GOOG_Stock_Price_Train.csv',index_col="Date",parse_dates=True)
print(df_train)
df_test = pd.read_csv(r'E:\工作\硕士\博客\博客37-\GOOG_Stock_Price_Test.csv',index_col="Date",parse_dates=True)
print(df_test )

输出结果：

            Open        High         Low       Close   Adj Close  \
Date                                                                     
2012-01-03  324.360352  331.916199  324.077179  330.555054  330.555054   
2012-01-04  330.366272  332.959412  328.175537  331.980774  331.980774   
2012-01-05  328.925659  329.839722  325.994720  327.375732  327.375732   
2012-01-06  327.445282  327.867523  322.795532  322.909790  322.909790   
2012-01-09  321.161163  321.409546  308.607819  309.218842  309.218842   
...                ...         ...         ...         ...         ...   
2016-12-23  790.900024  792.739990  787.280029  789.909973  789.909973   
2016-12-27  790.679993  797.859985  787.656982  791.549988  791.549988   
2016-12-28  793.700012  794.229980  783.200012  785.049988  785.049988   
2016-12-29  783.330017  785.929993  778.919983  782.789978  782.789978   
2016-12-30  782.750000  782.780029  770.409973  771.820007  771.820007   

              Volume  
Date                  
2012-01-03   7400800  
2012-01-04   5765200  
2012-01-05   6608400  
2012-01-06   5420700  
2012-01-09  11720900  
...              ...  
2016-12-23    623400  
2016-12-27    789100  
2016-12-28   1153800  
2016-12-29    742200  
2016-12-30   1770000  

[1258 rows x 6 columns]
                  Open        High         Low       Close   Adj Close  \
Date                                                                     
2017-01-03  778.809998  789.630005  775.799988  786.140015  786.140015   
2017-01-04  788.359985  791.340027  783.159973  786.900024  786.900024   
2017-01-05  786.080017  794.479980  785.020020  794.020020  794.020020   
2017-01-06  795.260010  807.900024  792.203979  806.150024  806.150024   
2017-01-09  806.400024  809.966003  802.830017  806.650024  806.650024   
...                ...         ...         ...         ...         ...   
2017-04-24  851.200012  863.450012  849.859985  862.760010  862.760010   
2017-04-25  865.000000  875.000000  862.809998  872.299988  872.299988   
2017-04-26  874.229980  876.049988  867.747986  871.729980  871.729980   
2017-04-27  873.599976  875.400024  870.380005  874.250000  874.250000   
2017-04-28  910.659973  916.849976  905.770020  905.960022  905.960022   

             Volume  
Date                 
2017-01-03  1657300  
2017-01-04  1073000  
2017-01-05  1335200  
2017-01-06  1640200  
2017-01-09  1272400  
...             ...  
2017-04-24  1372500  
2017-04-25  1672000  
2017-04-26  1237200  
2017-04-27  2026800  
2017-04-28  3219500  

[81 rows x 6 columns]

2、产生训练所需的特征和标签数据

X_train_set = df_train.iloc[:,4:5].values 
#数据归一化
sc = MinMaxScaler() 
X_train_set = sc.fit_transform(X_train_set)
 
def create_dataset(ds, look_back=1):
    X_data, Y_data = [],[]
    for i in range(len(ds)-look_back):
        X_data.append(ds[i:(i+look_back), 0])
        Y_data.append(ds[i+look_back, 0])
    return np.array(X_data), np.array(Y_data)
look_back = 60
print("回看天数:", look_back)
 
# 分割成特征数据和标签数据
X_train, Y_train = create_dataset(X_train_set, look_back)
 
X_train
Y_train

输出结果：

回看天数: 60

Out[5]:

array([0.08291369, 0.07626093, 0.0815312 , ..., 0.94758974, 0.94336851,
       0.92287887])

3、转换数据为（样本数，时步、特征）的张量

X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[1], 1))
X_train.shape 

输出结果：

(1198, 60, 1)

4、定义LSTM模型

      在编译模型中，损失函数为MSE，优化器为adam。在训练模型中，训练周期为100，批次尺寸为32。 

model = Sequential()
model.add(LSTM(50, return_sequences=True, 
               input_shape=(X_train.shape[1], 1)))
model.add(Dropout(0.2))
model.add(LSTM(50, return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(50))
model.add(Dropout(0.2))
model.add(Dense(1))
model.summary()  
#编译模型
model.compile(loss="mse", optimizer="adam") 
#训练模型
model.fit(X_train, Y_train, epochs=100, batch_size=32)

输出结果：

38/38 [==============================] - 2s 46ms/step - loss: 0.0013
Epoch 94/100
38/38 [==============================] - 2s 46ms/step - loss: 0.0013
Epoch 95/100
38/38 [==============================] - 2s 47ms/step - loss: 0.0012
Epoch 96/100
38/38 [==============================] - 2s 46ms/step - loss: 0.0013
Epoch 97/100
38/38 [==============================] - 2s 46ms/step - loss: 0.0013
Epoch 98/100
38/38 [==============================] - 2s 47ms/step - loss: 0.0013
Epoch 99/100
38/38 [==============================] - 2s 46ms/step - loss: 0.0012
Epoch 100/100
38/38 [==============================] - 2s 46ms/step - loss: 0.0013

5、使用已经训练好的LSTM模型预测股价

       测试集为2017年1月到3月的股价，因为使用的是前60天的股价数据，使用预测的是4月份股价 。

X_test_set = df_test.iloc[:,4:5].values
 
# 产生标签数据
_, Y_test = create_dataset(X_test_set, look_back)
 
#特征数据和标准化
X_test_s = sc.transform(X_test_set)
X_test,_ = create_dataset(X_test_s, look_back)
 
# 转换成(样本数, 时步, 特征)张量
X_test = np.reshape(X_test, (X_test.shape[0], X_test.shape[1], 1))
X_test_pred = model.predict(X_test)
 
#  将预测值转换回股价
X_test_pred_price = sc.inverse_transform(X_test_pred)
X_test_pred_price

输出结果：

array([[814.5596 ],
       [819.2384 ],
       [821.1239 ],
       [823.5624 ],
       [824.0013 ],
       [822.3476 ],
       [819.3523 ],
       [816.00055],
       [813.82117],
       [812.62726],
       [812.6262 ],
       [812.9471 ],
       [817.2544 ],
       [821.539  ],
       [824.44244],
       [826.5891 ],
       [828.0157 ],
       [834.4217 ],
       [843.3087 ],
       [849.4051 ],
       [852.694  ]], dtype=float32)

6、绘制真实股价与预测股价的对比图

import matplotlib.pyplot as plt
plt.plot(Y_test, color="red", label="Real Stock Price")
plt.plot(X_test_pred_price, color="blue", label="Predicted Stock Price")
plt.title("2017 Google Stock Price Prediction")
plt.xlabel("Time")
plt.ylabel("Google Time Price")
plt.legend()
plt.savefig("E:\工作\硕士\博客\博客37-/squares1.png",
            bbox_inches ="tight",
            pad_inches = 1,
            transparent = True,
            facecolor ="w",
            edgecolor ='w',
            dpi=300,
            orientation ='landscape')

输出结果：

---

更多优质内容持续发布中，请移步主页查看。

若有问题可邮箱联系：1736732074@qq.com 

博主的WeChat:TCB1736732074

   点赞+关注,下次不迷路！