【pytorch】构建多元时间序列数据集 Dataset

假设原本数据集是如下的 csv 格式，行代表时间，列数代表变量数。

用它来构造机器学习的数据集，也就是有监督标签的样本。

代码一

import torch
import torch.utils.data
import os
import numpy as np
import pandas as pd

class MTSDataset(torch.utils.data.Dataset):
    """Multi-variate Time-Series Dataset for *.txt file

    Returns:
        [sample, label]
    """

    def __init__(self,
                window,
                horizon,
                data_name='electricity',
                set_type='train',    # 'train'/'validation'/'test'
                data_dir='./data'):
        assert type(set_type) == type('str')
        self.window = window
        self.horizon = horizon
        self.data_dir = data_dir
        self.set_type = set_type

        file_path = os.path.join(data_dir, data_name, '{}_{}.txt'.format(data_name, set_type))

        rawdata = np.loadtxt(open(file_path), delimiter=',')
        self.len, self.var_num = rawdata.shape
        self.sample_num = max(self.len - self.window - self.horizon + 1, 0)
        self.samples, self.labels = self.__getsamples(rawdata)

    def __getsamples(self, data):
        X = torch.zeros((self.sample_num, self.window, self.var_num))
        Y = torch.zeros((self.sample_num, 1, self.var_num))

        for i in range(self.sample_num):
            start = i
            end = i + self.window
            X[i, :, :] = torch.from_numpy(data[start:end, :])
            Y[i, :, :] =