前言：

如果沒有深入了解LSTM原理及結構，推薦看下面兩篇blog，不在贅述：

從深度學習到LSTM：https://blog.csdn.net/hz371071798/article/details/82532183

LSTM結構詳解：https://blog.csdn.net/zhangbaoanhadoop/article/details/81952284

?

正文開始，簡單寫一下編程實現：

注：和上文一樣，data直接采用 ?facebook 的prophet時序算法中examples的數據。

一些引用的包：

            
              from __future__ import print_function
import numpy as np 
import pandas as pd 
from keras.layers import Dense, Activation, Dropout, LSTM
# from keras.layers.recurrent import LSTM
from keras.models import Sequential
from sklearn.model_selection  import train_test_split
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
import matplotlib.pyplot as plt
import math

import warnings
warnings.filterwarnings('ignore')
plt.style.use('seaborn-poster')
            
          

?

用到的函數：

            
              def create_dataset(dataset, look_back):
    '''
    # convert an array of values into a time series dataset
    :param dataset:
    :param look_back: step
    :return:
    '''
    dataX, dataY = [], []
    for i in range(len(dataset)-look_back-1):
        a = dataset[i:(i+look_back), 0]
        dataX.append(a)
        dataY.append(dataset[i + look_back, 0])
    return np.array(dataX), np.array(dataY)

def return_rmse(test,predicted):
    rmse = math.sqrt(mean_squared_error(test, predicted))
    print("rmse is {}.".format(rmse))
            
          

?

主函數：main_run

            
              # Load data
df = pd.read_csv('data/example_air_passengers.csv')
df.ds = pd.to_datetime(df.ds)
df.index = df.ds
df.drop(['ds'], axis=1, inplace=True)
print(df.head())
# print(df.info())
            
          

            
              # 基本參數設置，自行設定初始值：
look_back = 7
epochs = 1000
batch_size = 32
            
          

            
              # convert type
air_passengers_num = df.y.values.astype('float32')

# reshape to column vector
air_passengers_num = air_passengers_num.reshape(len(air_passengers_num), 1)
print(air_passengers_num)
            
          

正則化：

            
              # normalize
scaler = MinMaxScaler(feature_range=(0, 1))
air_passengers_num = scaler.fit_transform(air_passengers_num)
            
          

切分數據集并reshape

            
              # split data
train, test = air_passengers_num[0:train_size, :], air_passengers_num[train_size:len(air_passengers_num), :]

# split 
trainX, trainY = create_dataset(train, look_back)
testX, testY = create_dataset(test, look_back)

# reshape format
            
          

LSTM fit

            
              model = Sequential()
model.add(LSTM(4, input_shape=(look_back, 1)))
model.add(Dense(1))
model.compile(loss='mse', optimizer='adam')
model.fit(trainX, trainY, nb_epoch=epochs, batch_size=batch_size)
            
          

            
              # make predictions
trainPredict = model.predict(trainX)

# invert predictions and targets to unscaled
trainPredict = scaler.inverse_transform(trainPredict)
trainY = scaler.inverse_transform([trainY])


            
          

            
              # calculate rmse
trainScore = return_rmse(trainY[0], trainPredict[:, 0])
print('Train Score: {} RMSE'.format(trainScore))
testScore = return_rmse(testY[0], testPredict[:, 0])
print('Test Score: {} RMSE'.format(testScore))
            
          

plot

            
              # shift predictions of training data for plotting
trainPredictPlot = np.empty_like(air_passengers_num)
trainPredictPlot[:, :] = np.nan
trainPredictPlot[look_back:len(trainPredict) + look_back, :] = trainPred

# shift predictions of test data for plotting
testPredictPlot = np.empty_like(air_passengers_num)
testPredictPlot[:, :] = np.nan

# 這里留一個思考：look_back為什么*2
testPredictPlot[len(trainPredict) + (look_back*2) + 1:len(air_passengers_num) - 1, :] = testPred
# plot baseline and predictions
plt.plot(scaler.inverse_transform(air_passengers_num))
plt.plot(trainPredictPlot)
plt.plot(testPredictPlot)
plt.show()
            
          

            
              result：
Train Score: 24.93 RMSE
Test Score: 55.86 RMSE
            
          

?