An interface for multidimensional arrays in Arkouda

This extended abstract describes the prototype of a mul-tidimensional array interface that sits on top of Arkouda, a Python-based productivity layer for distributed array and dataframe computations [1] . Arkouda support has focused on 1-D arrays, offering an interface similar to Numpy (Python’s de facto standard sequential library for array processing) and using a Chapel-based backend for distributed execution. Motivated by tensor computations, we are exploring extensions of Arkouda to support multiway ("N-D") array objects. The goal of our prototype is not to claim a fully-featured, high-performance implementation but rather to initiate a broader discussion with the Arkouda community about whether and how to pursue such an extension. (For a practical, high-performance drop-in replacement for Numpy with distributed GPU support, see Legate [2] , among other efforts [3] – [6] .)