Uploaded image for project: 'SystemDS'
  1. SystemDS
  2. SYSTEMDS-3547

Tensor transposition on linearized data

    XMLWordPrintableJSON

Details

    • Task
    • Status: Open
    • Major
    • Resolution: Unresolved
    • None
    • None
    • None

    Description

      This project is to implement and extend the transpose operation to work on matrices via a shape and transpose order argument.
      The implementation should modify the matrices as if it has the tensor shape specified in the argument.
      A simple example of this is the python numpy transpose operator for multi dimensional matrices such as
      'X.transpose(2, 0, 1)' that change the order of the dimensions via a multi dimensional transpose.

      A suggested interface to implement would be :

      # read in matrix.
      # Shape is 2 by 12
      X = read(X)
      # A shape of 2 by 3 by 4 treating the matrix as a tensor with one extra dimension
      shape = matrix("2 3 4", rows=1, cols=3)
      # The order of the dimensions wanted after transpose
      order = matrix("1 3 2", rows = 1, cols=3)
      Xt, newShape = t(target=X, shape=shape, order=order)
      
      # Xt should now have shape 2 by 12 still, but
      # new shape is 2 by 4 by 3 
      print(toString(newShape))
      
      
      

      Attachments

        Activity

          People

            Unassigned Unassigned
            mboehm7 Matthias Boehm
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

            Dates

              Created:
              Updated: