[英]Is there a TensorFlow.js method to add a tensor to another at a specific location?
Let me make the question clear.让我把问题说清楚。 I have a tensor, just to keep it simple, with zeros.
我有一个张量,只是为了简单起见,带有零。 A 2D tensor with shape
[3, 3]
, call it a
:形状为
[3, 3]
二维张量,称为a
:
[[0,0,0],
[0,0,0],
[0,0,0]]
And I have another tensor, a 2D tensor with shape [2, 2]
, call it b
, and its just ones:我有另一个张量,一个形状为
[2, 2]
的二维张量,称为b
,它只是一个:
[[1,1],
[1,1]]
I want to add them together, but their shapes don't match, and I can't broadcast it as well, but it's not the goal.我想把它们加在一起,但它们的形状不匹配,我也无法播放,但这不是目标。 I want the result of:
我想要的结果是:
[[0,0,0],
[0,1,1],
[0,1,1]]
And if there would be a method, call it insert
for now, I'd have to specify the location where I want to insert the smaller tensor (just as when slicing, defining the start location).如果有一种方法,现在将其称为
insert
,我必须指定要插入较小张量的位置(就像切片时定义起始位置一样)。 It would look something like this:它看起来像这样:
a.insert(b, [1, 1])
I was thinking about pad
which could produce similar results, but if there would be something like the above, what I might missed, it would be nicer.我正在考虑可以产生类似结果的
pad
,但是如果有类似上面的东西,我可能会错过什么,那就更好了。
使用tf.pad
,可以按如下方式完成
padded = tf.ones([2, 2]).pad([[1, 0], [1, 0]])
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.