各个损失函数的计算公式,网上有很多文章了,此处就不一一介绍了。
多标签评价指标之Hamming Loss
PyTorch实现的Hamming Loss和sklearn实现的Hamming Loss代码实现对比
import torch
from sklearn.metrics import hamming_loss
def multi_label_classification_hamming_loss(preds, targets):
"""
计算多标签分类Hamming Loss的函数。
:param preds: 预测的概率值,大小为 [batch_size, num_classes]
:param targets: 目标标签值,大小为 [batch_size, num_classes]
:return: 多标签分类Hamming Loss的值,大小为 [1]
"""
# 将概率值转换为二进制标签(0或1)
binary_preds = torch.round(torch.sigmoid(preds))
# 计算Hamming Loss
hamming_loss = 1 - (binary_preds == targets).float().mean()
return hamming_loss
# 定义预测值和目标标签值
preds = torch.tensor([[0.1, 0.9, 0.3],
[0.8, 0.2, 0.6],
[0.4, 0.5, 0.7]])
targets = torch.tensor([[0, 1, 0],
[1, 0, 1],
[0, 1, 1]])
# 计算多标签分类Hamming Loss
loss = multi_label_classification_hamming_loss(preds, targets)
# 对比sklearn中的Hamming Loss计算结果
sklearn_loss = hamming_loss(targets.numpy(), torch.round(torch.sigmoid(preds)).numpy(), sample_weight=None)
print("PyTorch实现的Hamming Loss:", loss.item())
print("sklearn实现的Hamming Loss:", sklearn_loss)
输出结果:
PyTorch实现的Hamming Loss: 0.4444444179534912
sklearn实现的Hamming Loss: 0.4444444444444444
使用PyTorch中的torch.sigmo
6324



被折叠的 条评论
为什么被折叠?



