PyData/Sparse aims to replace scipy.sparse
as the choice for sparse arrays in the Python ecosystem.
Sparse is currently using the research done by the TACO team (https://github.com/tensor-compiler/taco)
to perform Sparse array computations efficiently. TACO has a Python API, called pytaco
, that makes the
C++ code available in Python by wrapping it up in pybind11. Sparse aims to use the pytaco
in a numpy.ndarray
compatible API.
My project roughly consists of three parts
- Complete revamp of the Sparse test suite by moving the tests to hypothesis, which uses property based testing.
- Update the cmake code of TACO so that it can be used by downstream users. This was mostly done by John Lee and is in this repository (https://github.com/leej3/taco/tree/cmake_changes).
- Creation of the
ndarray
compatible API mentioned. This is unfinished work in progress.
https://docs.google.com/document/d/1pH3h-TFY4s0BVA7sfC1bLPQDAmo4GUeDP7ovgkoSAs8/edit?usp=sharing