基本思路

• 所有的数据都叫Dataset
• 加载数据的叫DataLoader
• 每次加载多少条数据叫batch_size

所以构造训练数据的步骤是：首先把数据加载为Dataset，然后用DataLoader依次把数据传递到模型中即可

全部代码

import torch
import numpy as np
from torch.utils.data import Dataset, DataLoader


class MyDataset(Dataset):

    def __init__(self):
        txt_data = np.loadtxt('./my_data.txt', delimiter=',')
        self._x = torch.from_numpy(txt_data[:, :2])
        self._y = torch.from_numpy(txt_data[:, 2])
        self._len = len(txt_data)

    def __getitem__(self, item):  # 每次循环的时候返回的值
        return self._x[item], self._y[item]

    def __len__(self):
        return self._len


data = MyDataset()
dataloader = DataLoader(data, batch_size=3, shuffle=False, drop_last=True, num_workers=0)
n = 0
for x_data, y_label in dataloader:
    print('x:', x_data)
    print("y:", y_label)
    n += 1
print('迭代次数:', n)