pytorch筛选统计

sum是按我维度统计的，torch.sum 默认是所有求和，

import torch
if __name__ == '__main__':

    data=torch.linspace(1, 27, steps=27).view(9, 3)

    sum_a=sum(data>5)
    print(sum_a)

    print(torch.sum(data>5))

结果：

tensor([7, 7, 8])
tensor(22)

pytorch按维度统计：

import torch
if __name__ == '__main__':

    data=torch.linspace(1, 27, steps=27).view(9, 3)

    sum_a=sum(data>5)
    print(sum_a)

    print(data.shape)
    print(torch.sum(data>5,dim=0))
    print(torch.sum(data>5,dim=1))

结果：

tensor([7, 7, 8])
torch.Size([9, 3])
tensor([7, 7, 8])
tensor([0, 1, 3, 3, 3, 3, 3, 3, 3])

但是这个是不对的：

aa= torch.sigmoid(conf_preds[:, :, 0][pos_mask])

n_correct= int(sum(aa> 0.5)