今天，梅朵和大家分享用Python實現(xiàn)LSTM股價預(yù)測，僅用于技術(shù)學(xué)習(xí)交流，不作為投資建議。
1.導(dǎo)入依賴庫
主要是安裝相關(guān)的依賴庫。本文實現(xiàn)的環(huán)境為：python 3.7。
# 導(dǎo)入依賴庫import sklearnimport numpy as npimport pandas as pdimport matplotlib.pyplot as pltfrom keras.optimizers import Adamfrom keras.models import Sequentialfrom keras.layers import LSTM, Densefrom sklearn.preprocessing import MinMaxScaler2.讀取股價數(shù)據(jù)并歸一化
讀取股價數(shù)據(jù)文件，并對數(shù)據(jù)進(jìn)行歸一化處理。
# 讀取股票價格數(shù)據(jù)data = pd.read_csv('stock_data.csv')prices = data['Close'].values# 數(shù)據(jù)預(yù)處理：歸一化scaler = MinMaxScaler()prices = scaler.fit_transform(prices.reshape(-1, 1))
3.劃分訓(xùn)練集和測試集并確定模型輸入和輸出數(shù)據(jù)
劃分?jǐn)?shù)據(jù)的訓(xùn)練集和測試集，并確定模型輸入庫數(shù)據(jù)和輸出數(shù)據(jù)。
# 劃分?jǐn)?shù)據(jù)集為訓(xùn)練集和測試集 train_size = int(len(prices) * 0.75) train_data, test_data = prices[:train_size], prices[train_size:]
#確定模型輸入和輸出數(shù)據(jù) in_out_length = 12 train_sequences = [] for i in range(len(train_data) - in_out_length): train_sequences.append(train_data[i:i+in_out_length]) X_train = np.array(train_sequences) y_train = train_data[in_out_length:] test_sequences = [] for i in range(len(test_data) - in_out_length): test_sequences.append(test_data[i:i+in_out_length]) X_test = np.array(test_sequences) y_test = test_data[in_out_length:]4.創(chuàng)建LSTM模型并訓(xùn)練模型
創(chuàng)建LSTM模型并進(jìn)行模型訓(xùn)練。
# 創(chuàng)建LSTM模型model = Sequential()model.add(LSTM(50, input_shape=(in_out_length, 1)))model.add(Dense(1))learning_rate=0.002model.compile(loss='mean_squared_error', optimizer=Adam(learning_rate))# 訓(xùn)練模型model.fit(X_train, y_train, epochs=50, batch_size=64)
3.利用LSTM模型預(yù)測股價
利用訓(xùn)練好的LSTM模型進(jìn)行預(yù)測。
# 預(yù)測未來股價 predicted_prices = model.predict(X_test) predicted_prices = scaler.inverse_transform(predicted_prices) y_test = scaler.inverse_transform(y_test)完整源代碼
# 導(dǎo)入依賴庫import sklearnimport numpy as npimport pandas as pdimport matplotlib.pyplot as pltfrom keras.optimizers import Adamfrom keras.models import Sequentialfrom keras.layers import LSTM, Densefrom sklearn.preprocessing import MinMaxScalerif __name__=='__main__':# 讀取股票價格數(shù)據(jù)data = pd.read_csv('stock_data.csv')prices = data['Close'].values# 數(shù)據(jù)預(yù)處理：歸一化scaler = MinMaxScaler()prices = scaler.fit_transform(prices.reshape(-1, 1))# 劃分?jǐn)?shù)據(jù)集為訓(xùn)練集和測試集train_size = int(len(prices) * 0.75)train_data, test_data = prices[:train_size], prices[train_size:]#確定模型輸入和輸出數(shù)據(jù)in_out_length = 12train_sequences = []for i in range(len(train_data) - in_out_length):train_sequences.append(train_data[i:i+in_out_length])X_train = np.array(train_sequences)y_train = train_data[in_out_length:]test_sequences = []for i in range(len(test_data) - in_out_length):test_sequences.append(test_data[i:i+in_out_length])X_test = np.array(test_sequences)y_test = test_data[in_out_length:]# 創(chuàng)建LSTM模型model = Sequential()model.add(LSTM(50, input_shape=(in_out_length, 1)))model.add(Dense(1))learning_rate=0.002model.compile(loss='mean_squared_error', optimizer=Adam(learning_rate))# 訓(xùn)練模型model.fit(X_train, y_train, epochs=50, batch_size=64)# 預(yù)測未來股價predicted_prices = model.predict(X_test)predicted_prices = scaler.inverse_transform(predicted_prices)y_test = scaler.inverse_transform(y_test)# 繪制預(yù)測結(jié)果plt.plot(y_test, label='True_Prices')plt.plot(predicted_prices, label='Predicted_Prices')plt.legend()plt.xlabel('Time_Steps')plt.ylabel('Stock_Price')plt.title('LSTM Stock Price Prediction')plt.show()
