Pytorch采坑记录（持续更新）

1.将cuda tensor 转换成 ndarray:

loss.cpu().detach().numpy()

2.计算每个batch的准确率：

def accuary(output,labels):
    """
    获取每个batch的准确率
    """
    index = output.argmax(dim = 1) #获取dim=1维度上的最大值对应的索引组成的列表
    right_nums = 0
    all_num = len(labels)
    for i in range(all_num):
        if index[i] == labels[i]:
            right_nums += 1
    return round(right_nums/all_num,4)*100