【手写AI代码目录】准备发布的教程

1. tensorboard

from torch.utils.tensorboard import SummaryWriter

# TensorBoard
writer = SummaryWriter('runs/mnist_experiment_1')
...

        if i % 100 == 99:    # 每 100 个 batch 打印一次并记录到 TensorBoard
            print('[%d, %5d] loss: %.3f' %
                  (epoch + 1, i + 1, running_loss / 100))
            writer.add_scalar('Training Loss',
                              running_loss / 100,
                              epoch * len(trainloader) + i)
            writer.add_scalar('Accuracy',
                              correct / total,
                              epoch * len(trainloader) + i)
            running_loss = 0.0
            total = 0
            correct = 0
            
# 不要忘了关闭 SummaryWriter
writer.close()

然后终端启动命令： tensorboard --logdir=runs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

2. F.cross_entropy(input_tensor, target) = F.log_softmax() + F.nll_loss()

两者 计算的loss 相同

import torch
import torch.nn.functional as F
import torch.nn as nn

# 定义一个输入张量，形状为 (B, C)
input_tensor = torch.randn(3, 5)  # 假设批量大小为 3，类别数量为 5

target = torch.LongTensor([1, 0, 2])

# 使用 log_softmax 函数计算对数softmax值
output_tensor = F.log_softmax(input_tensor, dim=1)

# print("Input Tensor:")
# print(input_tensor)

print("\nOutput Tensor (Log-Softmax):")
print(output_tensor)

# 计算 NLL Loss
loss = F.nll_loss(output_tensor, target)
print("NLL Loss:", loss.item())

# 计算交叉熵损失
loss = F.cross_entropy(input_tensor, target)

print("Cross-Entropy Loss:", loss.item())

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27