Closed gbg141 closed 11 months ago
Check out this pull request on
See visual diffs & provide feedback on Jupyter Notebooks.
Powered by ReviewNB
All modified and coverable lines are covered by tests :white_check_mark:
Comparison is base (
2aaf18c
) 96.54% compared to head (642bdbf
) 97.49%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
This PR introduces some modifications in simplicial models to ensure a consistent implementation across all of them (with the exceptions of
SCA_CMPS
andSCConv
, that need further analysis).In particular, we decoupled the readout from the models, thus making a uniform implementation of the simplicial architectures for whichever dowstream task is considered. The output of each model is now the set of final hidden representations of the involved simplices (nodes, edgees, faces,... or a combination of them), and then a light
Network
class is defined in the tutorials to leverage the final hidden states to get the desired output for the considered dataset and task.Apart from this refactorization of the implementations, which could facilitate the aplication/adaptation of the models to different domains and tasks, we have also checked the model pipelines, as well as updated the involved test files accordingly (except with
SCA_CMPS
andSCConv
, which require a closer look).