Try   HackMD

Scientific Python Sparse Summit - Meeting 2

April 27th, 2023 - 11AM - 12PM PST

https://scientific-python.org/summits/sparse/meeting2/

Attendees

juanitagomezr2112@gmail.com
jim22k@gmail.com
dschult@colgate.edu
einstein.edison@gmail.com (Hameer Abbasi)
nabdennur@gmail.com
adam.li@columbia.edu
perimosocordiae@gmail.com (CJ)
erik.n.welch@gmail.com
Julien Jerphanion git@jjerphan.xyz

Who is going to be working on sparse arrays?

Julien Jerphanion
Ross Barnowski
Dan Schult
CJ Carey

Things to work before the summit

https://sparse.pydata.org/en/stable/
https://github.com/hameerabbasi/xsparse
https://github.com/Quansight-Labs/ndindex

PRs to be merged:

https://github.com/scipy/scipy/pull/16108
https://github.com/scipy/scipy/pull/16033

https://github.com/scipy/scipy/pull/16033#issuecomment-1120091359

Script:

A tool that understands the structure of code, and can make modifications. You can then ask to upgrade from scipy 0.x to scipy 0.y, and it would make some of the changes for you so you are not faced with deprecation warnings.

Example in NetworkX:
https://github.com/networkx/networkx/pull/5139

Some potential tools:
https://github.com/ssbr/refex
https://github.com/python-rope/rope
https://github.com/charliermarsh/ruff

Discussions

tl;dr Current problems with matrices:

  • Scipy sparse matrices convertion to dense array return numpy matrices

  • silently wrong answers with scipy matrices are a reason for removing this semantic

CJ Carey: not keen to vendoring numpy matrix within scipy, having a separate package for numpy matrice

Have sparse matrices by small wrappers over sparse arrays and then migrate

No sparse 1D array for now.

Road Map