Notes from GSOC mentor meetings with Kavish

July 18, 2023

Thoughts from research about parallel links:

not too many packages for network analysis on big parallel systems
most seem to be about using parallel on a local machine.
Might be good to start with functions that are already marked as embarrasingly parallelizable. This could help develop the framework with a very simple code change for going parallel.
There are many places in NX that basically takes a for-loop and computes the same thing for each of the nodes.
This posting has some interesting ideas: https://github.com/networkx/networkx/pull/6306

Todo:

work through the nx_parallel backend system
NX_parallel code… look at centrality.py and make these lines work~20-30
search through networkx for TODO and see which are already flagged as being good choices for parallel algorithms.

Links:

Larger scale plans:

12 week program ~175 hours.
Revised timeline which includes parallel work and Notebooks
For parallel work, start with betweenness and then BFS routines.
For notebooks, make one for Traversal
For parallel, has thought about using joblib or could work with dask – dask has a data structure called a Bag which could be used nicely for BFS.
We talked about nx_parallel and that brought up dispatching and backends. We talked about the big picture of how dispatching work. But we will need more discussion about that part of the process. Maybe a small team session focused on nx_parallel when Mridul is in the right timezone.

BLOG:

We talked about places to host the blog, and Kavish asked for an account on scientific-python to be able to post blogs there.

TODO:

[Dan] get an account set up on scientific-python.org
[Kavish] send the new timeline to Ross, Mridul and Dan
[Kavish] explore the blog site on scientific-python.org and start putting together the first blog.