第六讲 循环神经网络--LSTM--stock

  1 !pip install tushare
  2 import tushare as ts
  3 import numpy as np
  4 import tensorflow as tf
  5 from tensorflow.keras.layers import Dropout, Dense, LSTM
  6 import matplotlib.pyplot as plt
  7 import os
  8 import pandas as pd
  9 from sklearn.preprocessing import MinMaxScaler
 10 from sklearn.metrics import mean_squared_error, mean_absolute_error
 11 import math
 12 
 13 
 14 df1 = ts.get_k_data('600519', ktype='D', start='2004-01-01', end='2020-05-12')
 15 
 16 datapath1 = "./SH600519.csv"
 17 df1.to_csv(datapath1)
 18 
 19 
 20 maotai = pd.read_csv("./SH600519.csv")
 21 
 22 maotai.head()
 23 
 24 
 25 maotai.tail()
 26 
 27 
 28 training_set = maotai.iloc[0:3000, 2:3].values
 29 test_set = maotai.iloc[3000:, 2:3].values
 30 
 31 #归一化
 32 sc = MinMaxScaler(feature_range = (0, 1))
 33 training_set_scaled = sc.fit_transform(training_set)
 34 test_set = sc.transform(test_set)
 35 
 36 training_set_scaled.shape
 37 
 38 test_set.shape
 39 
 40 x_train = []
 41 y_train = []
 42 
 43 x_test = []
 44 y_test = []
 45 
 46 
 47 for i in range(60, len(training_set_scaled)):
 48   x_train.append(training_set_scaled[i - 60:i, 0])
 49   y_train.append(training_set_scaled[i, 0])
 50 
 51 np.random.seed(7)
 52 np.random.shuffle(x_train)
 53 np.random.seed(7)
 54 np.random.shuffle(y_train)
 55 tf.random.set_seed(7)
 56 
 57 
 58 x_train, y_train = np.array(x_train), np.array(y_train)
 59 
 60 x_train.shape
 61 y_train.shape
 62 
 63 
 64 x_train = np.reshape(x_train, (x_train.shape[0], 60, 1))
 65 for i in range(60, len(test_set)):
 66   x_test.append(test_set[i-60:i, 0])
 67   y_test.append(test_set[i, 0])
 68 
 69 x_test, y_test = np.array(x_test), np.array(y_test)
 70 x_test = np.reshape(x_test, (x_test.shape[0], 60, 1))
 71 
 72 
 73 model = tf.keras.Sequential([
 74         LSTM(80, return_sequences=True),
 75         Dropout(0.2),
 76         LSTM(100),
 77         Dropout(0.2),
 78         Dense(1)
 79 ])
 80 
 81 model.compile(optimizer=tf.keras.optimizers.Adam(0.0001),
 82               loss='mean_squared_error')
 83 
 84 checkpoint_save_path = "./checkpoint/LSTM_stock.ckpt"
 85 
 86 if os.path.exists(checkpoint_save_path + '.index'):
 87   print('-------------load the model-------------')
 88   model.load_weights(checkpoint_save_path)
 89 
 90 cp_callback = tf.keras.callbacks.ModelCheckpoint(
 91     filepath=checkpoint_save_path,
 92     save_weights_only=True,
 93     save_best_only=True,
 94     monitor='val_loss')
 95 
 96 history = model.fit(x_train, y_train, batch_size=64, epochs=24, 
 97                     validation_data=(x_test, y_test), validation_freq=1, callbacks=[cp_callback])
 98 
 99 model.summary()
100 
101 
102 
103 with open("./weights.txt", "w") as f:
104   for v in model.trainable_variables:
105     f.write(str(v.name) + '\n')
106     f.write(str(v.shape) + '\n')
107     f.write(str(v.numpy()) + '\n')
108 
109 
110 loss = history.history['loss']
111 val_loss = history.history['val_loss']
112 
113 plt.plot(loss, label='Training Loss')
114 plt.plot(val_loss, label='Validation Loss')
115 plt.title('Training and Validation Loss')
116 plt.legend()
117 plt.show()
118 
119 
120 predicted_stock_price = model.predict(x_test)
121 predicted_stock_price = sc.inverse_transform(predicted_stock_price)
122 real_stock_price = sc.inverse_transform(test_set[60:])
123 
124 plt.plot(real_stock_price, color='red', label='real_stock_price')
125 plt.plot(predicted_stock_price, color='blue', label='predicted_stock_price')
126 plt.title('Maotai Stock Price Prediction')
127 plt.xlabel('Time')
128 plt.ylabel('Maotai Stock Price')
129 plt.legend()
130 plt.show()
131 
132 
133 mse = mean_squared_error(predicted_stock_price, real_stock_price)
134 rmse = math.sqrt(mean_squared_error(predicted_stock_price, real_stock_price))
135 mae = mean_absolute_error(predicted_stock_price, real_stock_price)
136 print('均方误差： %.6f'%mse)
137 print('均方根误差： %.6f'%rmse)
138 print('平均绝对误差： %.6f'%mae)