Python 火炬和沿轴的张量
声明：本页面是StackOverFlow热门问题的中英对照翻译，遵循CC BY-SA 4.0协议，如果您需要使用它，必须同样遵循CC BY-SA许可，注明原文地址和作者信息，同时你必须将它归于原作者（不是我）：StackOverFlow
原文地址: http://stackoverflow.com/questions/44790670/
Warning: these are provided under cc-by-sa 4.0 license. You are free to use/share it, But you must attribute it to the original authors (not me):
StackOverFlow
Torch sum a tensor along an axis
提问by Abhishek Bhatia
ipdb> outputs.size()
torch.Size([10, 100])
ipdb> print sum(outputs,0).size(),sum(outputs,1).size(),sum(outputs,2).size()
(100L,) (100L,) (100L,)
How do I sum over the columns instead?
我如何对列求和？
回答by mexmex
The simplest and best solution is to use torch.sum().
最简单和最好的解决方案是使用torch.sum().
To sum all elements of a tensor:
对张量的所有元素求和：
torch.sum(outputs) # gives back a scalar
To sum over all rows (i.e. for each column):
对所有行（即每列）求和：
torch.sum(outputs, dim=0) # size = [1, ncol]
To sum over all columns (i.e. for each row):
对所有列（即每一行）求和：
torch.sum(outputs, dim=1) # size = [nrow, 1]
回答by kmario23
Alternatively, you can use tensor.sum(axis)where axisindicates 0and 1for summing over rows and columns respectively, for a 2D tensor.
或者，对于二维张量，您可以使用tensor.sum(axis)whereaxis指示0和1分别对行和列求和。
In [210]: X
Out[210]:
tensor([[ 1, -3, 0, 10],
[ 9, 3, 2, 10],
[ 0, 3, -12, 32]])
In [211]: X.sum(1)
Out[211]: tensor([ 8, 24, 23])
In [212]: X.sum(0)
Out[212]: tensor([ 10, 3, -10, 52])
As, we can see from the above outputs, in both cases, the output is a 1D tensor. If you, on the other hand, wish to retain the dimension of the original tensor in the output as well, then you've set the boolean kwarg keepdimto Trueas in:
从上面的输出我们可以看出，在这两种情况下，输出都是一维张量。另一方面，如果您还希望在输出中保留原始张量的维度，那么您已将布尔 kwargkeepdim设置True为：
In [217]: X.sum(0, keepdim=True)
Out[217]: tensor([[ 10, 3, -10, 52]])
In [218]: X.sum(1, keepdim=True)
Out[218]:
tensor([[ 8],
[24],
[23]])
回答by postylem
If you have tensor my_tensor, and you wish to sum across the second array dimension (that is, the one with index 1, which is the column-dimension, if the tensor is 2-dimensional, as yours is), use torch.sum(my_tensor,1)or equivalently my_tensor.sum(1)see documentation here.
如果您有 tensor my_tensor，并且您希望对第二个数组维度求和（即，索引为 1 的那个维度，即列维度，如果张量是二维的，就像您的一样），请使用torch.sum(my_tensor,1)或等效地my_tensor.sum(1)参见文档在这里。
One thing that is not mentioned explicitly in the documentation is: you can sum across the lastarray-dimension by using -1(or the second-to last dimension, with -2, etc.)
文档中没有明确提到的一件事是：您可以通过使用（或倒数第二个维度，with等）对最后一个数组维度求和-1-2
So, in your example, you could use: outputs.sum(1)or torch.sum(outputs,1), or, equivalently, outputs.sum(-1)or torch.sum(outputs,-1). All of these would give the same result, an output tensor of size torch.Size([10]), with each entry being the sum over the all rows in a given column of the tensor outputs.
因此，在您的示例中，您可以使用：outputs.sum(1)或torch.sum(outputs,1)，或，等价地， outputs.sum(-1)或torch.sum(outputs,-1)。所有这些都会给出相同的结果，一个大小为 的输出张量torch.Size([10])，每个条目是 tensor 给定列中所有行的总和outputs。
To illustrate with a 3-dimensional tensor:
用 3 维张量来说明：
In [1]: my_tensor = torch.arange(24).view(2, 3, 4)
Out[1]:
tensor([[[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]],
[[12, 13, 14, 15],
[16, 17, 18, 19],
[20, 21, 22, 23]]])
In [2]: my_tensor.sum(2)
Out[2]:
tensor([[ 6, 22, 38],
[54, 70, 86]])
In [3]: my_tensor.sum(-1)
Out[3]:
tensor([[ 6, 22, 38],
[54, 70, 86]])
