PyTorch 笔记（07）— Tensor 的归并运算（torch.mean、sum、median、mode、norm、dist、std、var、cumsum、cumprod）
1. Tensor 归并运算函数
此类操作会使输出形状小于输入形状，并可以沿着某一维度进行指定操作，如加法， 既可以计算整个 tensor 的和，也可以计算 tensor 每一行或者 每一列的和，
常用归并操作如下表所示：
2. 使用示例
2.1 torch.mean
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [13]: t.mean(a)
Out[13]: tensor(2.5000)
In [14]: t.mean(a,dim=0)
Out[14]: tensor([1.5000, 3.5000])
In [15]: t.mean(a,dim=1)
Out[15]: tensor([3., 2.])
2.2 torch.sum
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [16]: a.sum()
Out[16]: tensor(10.)
In [17]: a.sum(dim=0)
Out[17]: tensor([3., 7.])
In [18]: a.sum(dim=1)
Out[18]: tensor([6., 4.])
In [19]:
2.3 torch.median
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [19]: a.median()
Out[19]: tensor(2.)
In [20]: a.median(dim=0)
Out[20]:
torch.return_types.median(
values=tensor([1., 3.]),
indices=tensor([1, 1]))
In [21]: a.median(dim=1)
Out[21]:
torch.return_types.median(
values=tensor([2., 1.]),
indices=tensor([0, 0]))
2.4 torch.mode
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [29]: a.mode()
Out[29]:
torch.return_types.mode(
values=tensor([2., 1.]),
indices=tensor([0, 0]))
In [30]: a.mode(dim=0)
Out[30]:
torch.return_types.mode(
values=tensor([1., 3.]),
indices=tensor([1, 1]))
In [31]: a.mode(dim=1)
Out[31]:
torch.return_types.mode(
values=tensor([2., 1.]),
indices=tensor([0, 0]))
2.5 torch.norm
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [22]: a.norm()
Out[22]: tensor(5.4772)
In [23]: a.norm(dim=0)
Out[23]: tensor([2.2361, 5.0000])
In [24]: a.norm(dim=1)
Out[24]: tensor([4.4721, 3.1623])
2.6 torch.dist
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [28]: a.dist(t.Tensor([1,2]))
Out[28]: tensor(2.4495)
2.7 torch.std
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [32]: a.std()
Out[32]: tensor(1.2910)
In [33]: a.std(dim=0)
Out[33]: tensor([0.7071, 0.7071])
In [34]: a.std(dim=1)
Out[34]: tensor([1.4142, 1.4142])
2.8 torch.var
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [35]: a.var()
Out[35]: tensor(1.6667)
In [36]: a.var(dim=0)
Out[36]: tensor([0.5000, 0.5000])
In [37]: a.var(dim=1)
Out[37]: tensor([2., 2.])
2.9 torch.cumsum
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [39]: a.cumsum(dim=0)
Out[39]:
tensor([[2., 4.],
[3., 7.]])
In [40]: a.cumsum(dim=1)
Out[40]:
tensor([[2., 6.],
[1., 4.]])
2.10 torch.cumprod
In [1]: import torch as t
In [11]: a = t.Tensor([[2,4], [1, 3]])
In [12]: a
Out[12]:
tensor([[2., 4.],
[1., 3.]])
In [41]: a.cumprod(dim=0)
Out[41]:
tensor([[ 2., 4.],
[ 2., 12.]])
In [42]: a.cumprod(dim=1)
Out[42]:
tensor([[2., 8.],
[1., 3.]])