koki watanabe
koki watanabe
> - We need the docs to have a reference for Kunegis et al (2010). > - We should get a better name than use_abs. What should this variant be...
@fernandodiazdiaz @dschult As part of this PR, would it make sense to rename the current default variant to "unsigned" and use_abs to "opposing", and additionally introduce a new "repelling" variant?...
@fernandodiazdiaz @dschult I’ve gone through Shi, Altafini & Baras (2019) “Dynamics over Signed Networks” to make sure I fully understood the mathematical distinctions. From that reading it’s clear that: -...
fixed some tests and docs
Addresses an environment-dependent bug.
> For the "unsigned" variant we replace the edge weights with their absolute value. And in some places, the PR overwrites the values in the graph object to avoid memory...
Thank you for catching this. I’ve updated the tests to validate the behavior of the copy parameter when edge weights are negative, and I’ve also corrected the laplacian implementation accordingly.
@dschult Thank you for the feedback! Based on your suggestions, I’ve renamed the parameter to signed_graph_variant (defaulting to standard), updated the docstring to include the explicit formulas for each variant,...
~The remaining failing test is the issue of #23412 so not related to this PR.~ This seems to have been fixed in #23381
@j-bowhay Thank you for pointing that out! I’ve added handling for sparse matrices in both DOK and LIL formats.