Consider:
T = torch.randn(u,v,w,x,y,z)
where u,v,w,x,y,z are some dimensions. If I want to reshape this as: u,v,w,x,y*z I can write:
T = torch.randn(u,v,w,x,y,z).view(-1,v,w,x,y*z)
But I cannot write:
T = torch.randn(u,v,w,x,y,z).view(-1,y*z)
Since this will flatten all previous dimensions.
Is there some convenient notation like:
T = torch.randn(u,v,w,x,y,z).view_last(y*z)
U = T.view_last(y,z)
Currently in PyTorch, that just reshapes the final dimensions?
The reason I am interested in this is the case where u,v,w,x are not known ahead of time, and I would rather avoid doing a T.shape.