I want to find a pytorch function which can do the operation as follows:
a=[ [1 2 3], [1,3,4], [2,5,7] ] #a.shape is 3*3
I want to obttain a expanded tensor as follows:
b=[ [1,2,3], [0,0,0], [1,3,4], [2,5,7], [0,0,0] ] #b.shape is 5*3
What I want to do is to expand the tensor a
to a larger tensor b
and assign the selected tensor index such as [0,2,3]
from a
, and fill other index [1,4]
with zero.
I have tried to define a large tensor and use torch.tensor.index_copy_ to do this operation, but this need to define a extral tensor, which brings additional storage, can anyone give some other methods to do this?