批数据训练

批数据训练

拥有大量数据时，可把数据分批送入神经网络训练。

分批代码如下：

import torch
import torch.utils.data as Data

if __name__ == '__main__':

    BATCH_SIZE = 5

    x = torch.linspace(1, 10, 10)
    y = torch.linspace(10, 1, 10)

    torch_dataset = Data.TensorDataset(x, y)
    loader = Data.DataLoader(
        dataset=torch_dataset,
        batch_size=BATCH_SIZE,
        shuffle=True,  # shuffle the data
        num_workers=2,  # by 2 threads
    )

    for epoch in range(3):  # train the whole data 3 times
        for step, (batch_x, batch_y) in enumerate(loader):  # enumerate:add index to elements in loader

            # fake training
            print('Epoch: ', epoch, '| Step: ', step, '| batch x: ',
                  batch_x.numpy(), '| batch y: ', batch_y.numpy())

输出结果：

Epoch:  0 | Step:  0 | batch x:  [3. 1. 9. 6. 2.] | batch y:  [ 8. 10.  2.  5.  9.]
Epoch:  0 | Step:  1 | batch x:  [10.  8.  5.  4.  7.] | batch y:  [1. 3. 6. 7. 4.]
Epoch:  1 | Step:  0 | batch x:  [ 4.  2. 10.  6.  9.] | batch y:  [7. 9. 1. 5. 2.]
Epoch:  1 | Step:  1 | batch x:  [3. 8. 5. 1. 7.] | batch y:  [ 8.  3.  6. 10.  4.]
Epoch:  2 | Step:  0 | batch x:  [4. 8. 2. 6. 7.] | batch y:  [7. 3. 9. 5. 4.]
Epoch:  2 | Step:  1 | batch x:  [ 3. 10.  9.  1.  5.] | batch y:  [ 8.  1.  2. 10.  6.]

一共有10对数据，批大小BATCH_SIZE=5，因此分为10/5=2批数据，每次完整的训练需要2步，总共进行了3次完整训练。

若BATCH_SIZE=8，则输出结果为：

Epoch:  0 | Step:  0 | batch x:  [ 4. 10.  2.  6.  5.  7.  9.  1.] | batch y:  [ 7.  1.  9.  5.  6.  4.  2. 10.]
Epoch:  0 | Step:  1 | batch x:  [8. 3.] | batch y:  [3. 8.]
Epoch:  1 | Step:  0 | batch x:  [10.  1.  4.  7.  8.  2.  5.  3.] | batch y:  [ 1. 10.  7.  4.  3.  9.  6.  8.]
Epoch:  1 | Step:  1 | batch x:  [9. 6.] | batch y:  [2. 5.]
Epoch:  2 | Step:  0 | batch x:  [10.  4.  8.  7.  5.  6.  3.  9.] | batch y:  [1. 7. 3. 4. 6. 5. 8. 2.]
Epoch:  2 | Step:  1 | batch x:  [1. 2.] | batch y:  [10.  9.]

依然是10对数据，批大小BATCH_SIZE=8，因此分为10/8=1余2批数据，1余2取上界=2批，第一批大小=8，第二批凑不满8则取剩下的2个数据，因此每次完整的训练需要2步，总共进行了3次完整训练。