pytorch lstm 预测航空旅客数目

airflights passengers dataset下载地址https://raw.githubusercontent.com/jbrownlee/Datasets/master/airline-passengers.csv

这个dataset包含从1949年到1960年每个月的航空旅客数目，共12*12=144个数字。

下面的程序中，我们以1949-1952的数据预测1953的数据，以1950-1953的数据预测1954的数据，以此类推，训练模型。

100

101

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import torch
import torch.nn as nn
from sklearn.preprocessing import MinMaxScaler
import os
 
# super parameters
EPOCH = 400
learning_rate = 0.01
seq_length = 4   # 序列长度
n_feature = 12   # 序列中每个元素的特征数目。本程序采用的序列元素为一年的旅客，一年12个月，即12维特征。
 
# data
data = pd.read_csv('airline-passengers.csv')   # 共 "12年*12个月=144" 个数据
data = data.iloc[:, 1:5].values        # dataFrame, shape (144,1)
data = np.array(data).astype(np.float32)
sc = MinMaxScaler()
data = sc.fit_transform(data)          # 归一化
data = data.reshape(-1, n_feature)     # shape (12, 12)
 
trainData_x = []
trainData_y = []
for i in range(data.shape[0]-seq_length):
    tmp_x = data[i:i+seq_length, :]
    tmp_y = data[i+seq_length, :]
    trainData_x.append(tmp_x)
    trainData_y.append(tmp_y)
 
# model
class Net(nn.Module):
    def __init__(self, in_dim=12, hidden_dim=10, output_dim=12, n_layer=1):
        super(Net, self).__init__()
        self.in_dim = in_dim
        self.hidden_dim = hidden_dim
        self.output_dim = output_dim
        self.n_layer = n_layer
        self.lstm = nn.LSTM(input_size=in_dim, hidden_size=hidden_dim, num_layers=n_layer, batch_first=True)
        self.linear = nn.Linear(hidden_dim, output_dim)
 
    def forward(self, x):
        _, (h_out, _) = self.lstm(x)  # h_out是序列最后一个元素的hidden state
                                      # h_out's shape (batchsize, n_layer*n_direction, hidden_dim), i.e. (1, 1, 10)
                                      # n_direction根据是“否为双向”取值为1或2
        h_out = h_out.view(h_out.shape[0], -1)   # h_out's shape (batchsize, n_layer * n_direction * hidden_dim), i.e. (1, 10)
        h_out = self.linear(h_out)    # h_out's shape (batchsize, output_dim), (1, 12)
        return h_out
 
train = True
if train:
    model = Net()
    loss_func = torch.nn.MSELoss()
    optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
    # train
    for epoch in range(EPOCH):
        total_loss = 0
        for iteration, X in enumerate(trainData_x):  # X's shape (seq_length, n_feature)
            X = torch.tensor(X).float()
            X = torch.unsqueeze(X, 0)                # X's shape (1, seq_length, n_feature), 1 is batchsize
            output = model(X)       # output's shape (1,12)
            output = torch.squeeze(output)
            loss = loss_func(output, torch.tensor(trainData_y[iteration]))
            optimizer.zero_grad()   # clear gradients for this training iteration
            loss.backward()         # computing gradients
            optimizer.step()        # update weights
            total_loss += loss
 
        if (epoch+1) % 20 == 0:
            print('epoch:{:3d}, loss:{:6.4f}'.format(epoch+1, total_loss.data.numpy()))
    # torch.save(model, 'flight_model.pkl')  # 这样保存会弹出UserWarning，建议采用下面的保存方法，详情可参考https://zhuanlan.zhihu.com/p/129948825
    torch.save({'state_dict': model.state_dict()}, 'checkpoint.pth.tar')
 
else:
    # model = torch.load('flight_model.pth')
    model = Net()
    checkpoint = torch.load('checkpoint.pth.tar')
    model.load_state_dict(checkpoint['state_dict'])
 
# predict
model.eval()
predict = []
for X in trainData_x:             # X's shape (seq_length, n_feature)
    X = torch.tensor(X).float()
    X = torch.unsqueeze(X, 0)     # X's shape (1, seq_length, n_feature), 1 is batchsize
    output = model(X)             # output's shape (1,12)
    output = torch.squeeze(output)
    predict.append(output.data.numpy())
 
# plot
plt.figure()
predict = np.array(predict)
predict = predict.reshape(-1, 1).squeeze()
x_tick = np.arange(len(predict)) + (seq_length*n_feature)
plt.plot(list(x_tick), predict, label='predict data')
 
data_original = data.reshape(-1, 1).squeeze()
plt.plot(range(len(data_original)), data_original, label='original data')
 
plt.legend(loc='best')
plt.show()