Suppose you had sample = tensor([[3,5,4]   [0,2,1]])

Then these will all return the exact same output:

• sample.repeat(2,1)
• sample.view(1,-1).expand(2,-1).contiguous().view(4,3)
• sample.index_select(0, tensor([0,1,0,1])
• torch. cat( [sample, sample], 0 )

Explanations

• Repeat - this is the correct tool for the job
• Expand - meant for expanding a tensor when one of the dimensions is a singleton
  • in other words, if the sample is (4,1) and we want to repeat (4,3)
  • we should use sample.expand(4,3) and not sample.repeat(1,3)
• Index Select - meant for re-ordering the items in a tensor
  • so we might have a tensor that was shuffled, and we want to shuffle it back into place
• Concat - meant for joining together two different tensors
  • also, do not confuse with torch.stack, which would add an extra dimension
  • concat a list of four 2x3 matrices and you will get 8x3 back
  • stack a list of four 2x3 matrices and you will get 4x2x3 back