前言
近期回顾LSTM 做时间序列数据预测,网上也有很多的教程,在跑这个程序时,遇到一些问题,特此记录分享一下。 使用LSTM 进行单步预测和多步预测,LSTM 的输出格式要重新调整,简单演示,不调参数。
参考文章: Time Series Prediction with LSTM Recurrent Neural Networks in Python with Keras 外国学者写的文章,比较具体,有兴趣可以进去看看。
1. 数据集获取
链接: https://pan.baidu.com/s/1jv7A2JvIhA6oqvtYnYh9vQ 提取码: m5j5
2. 模型实验
数据展示  1949 到 1960 一共 12 年,每年 12 个月的数据,一共 144 个数据,单位是 1000
2.1 单步预测
用当前数据预测下一个数据 目标:预测国际航班未来 1 个月的乘客数
import numpy
import matplotlib.pyplot as plt
from pandas import read_csv
import math
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
dataframe = read_csv('airline-passengers.csv', usecols=[1], engine='python')
print("数据集的长度:",len(dataframe))
dataset = dataframe.values
dataset = dataset.astype('float32')
plt.plot(dataset)
plt.show()

- 数据格式转换为监督学习,归一化数据,训练集和测试集划分
def create_dataset(dataset, look_back=1):
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 numpy.array(dataX), numpy.array(dataY)
numpy.random.seed(7)
scaler = MinMaxScaler(feature_range=(0, 1))
dataset = scaler.fit_transform(dataset)
train_size = int(len(dataset) * 0.67)
test_size = len(dataset) - train_size
train, test = dataset[0:train_size,:], dataset[train_size:len(dataset),:]
print("原始训练集的长度:",train_size)
print("原始测试集的长度:",test_size)
- 构建模型、预测
训练集长度96 ,监督学习之后变为94 原因:一是因为第96个数据没有预测值,这个值忽略;而是因为构建监督学习时,代码自动过滤一个值。所以变成94个数据。代码那块也可以不自动过滤一个值,对代码修改一下也行。
我当时在这里耗费很大的精力
look_back = 1
trainX, trainY = create_dataset(train, look_back)
testX, testY = create_dataset(test, look_back)
print("转为监督学习,训练集数据长度:", len(trainX))
print("转为监督学习,测试集数据长度:",len(testX))
trainX = numpy.reshape(trainX, (trainX.shape[0], 1, trainX.shape[1]))
testX = numpy.reshape(testX, (testX.shape[0], 1, testX.shape[1]))
print('构造得到模型的输入数据(训练数据已有标签trainY): ',trainX.shape,testX.shape)
model = Sequential()
model.add(LSTM(4, input_shape=(1, look_back)))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='adam')
model.fit(trainX, trainY, epochs=100, batch_size=1, verbose=2)
model.summary()
trainPredict = model.predict(trainX)
testPredict = model.predict(testX)
trainPredict = scaler.inverse_transform(trainPredict)
trainY = scaler.inverse_transform([trainY])
testPredict = scaler.inverse_transform(testPredict)
testY = scaler.inverse_transform([testY])
trainScore = math.sqrt(mean_squared_error(trainY[0], trainPredict[:,0]))
print('Train Score: %.2f RMSE' % (trainScore))
testScore = math.sqrt(mean_squared_error(testY[0], testPredict[:,0]))
print('Test Score: %.2f RMSE' % (testScore))
trainPredictPlot = numpy.empty_like(dataset)
trainPredictPlot[:, :] = numpy.nan
trainPredictPlot[look_back:len(trainPredict)+look_back, :] = trainPredict
testPredictPlot = numpy.empty_like(dataset)
testPredictPlot[:, :] = numpy.nan
testPredictPlot[len(trainPredict)+(look_back*2)+1:len(dataset)-1, :] = testPredict
plt.plot(scaler.inverse_transform(dataset))
plt.plot(trainPredictPlot)
plt.plot(testPredictPlot)
plt.show()
这里说下LSTM的输入格式为:samples, time steps, features input_shape的格式为:time steps, features 这两个地方容易出错
原始训练集的长度: 96
原始测试集的长度: 48
转为监督学习,训练集数据长度: 94
转为监督学习,测试集数据长度: 46
构造得到模型的输入数据(训练数据已有标签trainY): (94, 1, 1) (46, 1, 1)
Epoch 1/100
此过程省略
Model: "sequential_49"
_________________________________________________________________
Layer (type) Output Shape Param
=================================================================
lstm_49 (LSTM) (None, 4) 96
_________________________________________________________________
dense_49 (Dense) (None, 1) 5
=================================================================
Total params: 101
Trainable params: 101
Non-trainable params: 0
_________________________________________________________________
Train Score: 22.62 RMSE
Test Score: 51.93 RMSE

finalX = numpy.reshape(test[-1], (1, 1, testX.shape[1]))
featruePredict = model.predict(finalX)
featruePredict = scaler.inverse_transform(featruePredict)
print('模型预测1961年1月份的国际航班人数是: ',featruePredict)
结果: 模型预测1961年1月份的国际航班人数是: [[252.41042]]
2.2 多步预测
前三步预测下一步数据
import numpy
import matplotlib.pyplot as plt
from pandas import read_csv
import math
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
dataframe = read_csv('airline-passengers.csv', usecols=[1], engine='python')
print("数据集的长度:",len(dataframe))
dataset = dataframe.values
dataset = dataset.astype('float32')
plt.plot(dataset)
plt.show()
def create_dataset(dataset, look_back=1):
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 numpy.array(dataX), numpy.array(dataY)
numpy.random.seed(7)
scaler = MinMaxScaler(feature_range=(0, 1))
dataset = scaler.fit_transform(dataset)
train_size = int(len(dataset) * 0.67)
test_size = len(dataset) - train_size
train, test = dataset[0:train_size,:], dataset[train_size:len(dataset),:]
print("原始训练集的长度:",train_size)
print("原始测试集的长度:",test_size)
look_back = 3
trainX, trainY = create_dataset(train, look_back)
testX, testY = create_dataset(test, look_back)
print("转为监督学习,训练集数据长度:", len(trainX))
print("转为监督学习,测试集数据长度:",len(testX))
trainX = numpy.reshape(trainX, (trainX.shape[0], trainX.shape[1],1))
testX = numpy.reshape(testX, (testX.shape[0], testX.shape[1], 1))
print('构造得到模型的输入数据(训练数据已有标签trainY): ',trainX.shape,testX.shape)
model = Sequential()
model.add(LSTM(4, input_shape=( look_back,1)))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='adam')
model.fit(trainX, trainY, epochs=100, batch_size=1, verbose=2)
model.summary()
trainPredict = model.predict(trainX)
testPredict = model.predict(testX)
trainPredict = scaler.inverse_transform(trainPredict)
trainY = scaler.inverse_transform([trainY])
testPredict = scaler.inverse_transform(testPredict)
testY = scaler.inverse_transform([testY])
trainScore = math.sqrt(mean_squared_error(trainY[0], trainPredict[:,0]))
print('Train Score: %.2f RMSE' % (trainScore))
testScore = math.sqrt(mean_squared_error(testY[0], testPredict[:,0]))
print('Test Score: %.2f RMSE' % (testScore))
trainPredictPlot = numpy.empty_like(dataset)
trainPredictPlot[:, :] = numpy.nan
trainPredictPlot[look_back:len(trainPredict)+look_back, :] = trainPredict
testPredictPlot = numpy.empty_like(dataset)
testPredictPlot[:, :] = numpy.nan
testPredictPlot[len(trainPredict)+(look_back*2)+1:len(dataset)-1, :] = testPredict
plt.plot(scaler.inverse_transform(dataset))
plt.plot(trainPredictPlot)
plt.plot(testPredictPlot)
plt.show()
原始训练集的长度: 96
原始测试集的长度: 48
转为监督学习,训练集数据长度: 92
转为监督学习,测试集数据长度: 44
构造得到模型的输入数据(训练数据已有标签trainY): (92, 3, 1) (44, 3, 1)
Epoch 1/100
92/92 - 2s - loss: 0.0203
此过程省略
Model: "sequential_48"
_________________________________________________________________
Layer (type) Output Shape Param
=================================================================
lstm_48 (LSTM) (None, 4) 96
_________________________________________________________________
dense_48 (Dense) (None, 1) 5
=================================================================
Total params: 101
Trainable params: 101
Non-trainable params: 0
_________________________________________________________________
Train Score: 22.11 RMSE
Test Score: 50.08 RMSE

finalX = numpy.reshape(test[-3:], (1, testX.shape[1], 1))
featruePredict = model.predict(finalX)
featruePredict = scaler.inverse_transform(featruePredict)
print('模型预测1961年1月份的国际航班人数是: ',featruePredict)
模型预测1961年1月份的国际航班人数是: [[461.36993]]
不要在意结果,你懂的!!!
3. 总结
- 使用LSTM网络,数据要归一化处理, 格式要处理
- 输入格式:samples, time steps, features(维度)
input_shape的格式为:time steps, features - LSTM内部超参数可以调节,层数可以增加
- 从数据的角度,考虑数据预处理的情况,比如差值填充之类的,这里没有用到。
- 在划分数据集时,训练集长度,转为监督学习后长度会缩小。
有个疑问: 多步预测时
这个三个部分对应修改,程序也可以正常运行,预测结果和上述也差不多。 上面肯定是对的,这里的处理,我认为是数据处理的格式不同;比如说上面是前三步预测下一步,这里处理的话就是前一步预测下三步,模型都能跑通。 有大佬走过路过的话,烦请解答一下,万分感激。
|