TopoTune

[mathildepapillon]

TopoTune : Capability for creating graph-based TNNs using GCCN models

This pull request encapsulates all necessary files and scripts for running TopoTune within TopoBenchamrkX, a framework for defining and training a wide variety of graph-based TNNs. The framework is comprised of

• combinatorial model TopoTune (in gccn.py), which expands an input complex into an ensemble of strictly augmented Hasse graphs and processes them with a sequence of synchronized base neural networks, such as a GNNs or Transformers
• combinatorial model TopoTune_OneHasse (in gccn_onehasse.py), which expands an input complex into a single augmented Hasse graph and processes it with a sequence of a base neural network, such as a GNN or Transformer
• configuration files for running a wide variety of these models in the simplicial and cellular domains
• configuration files for reproducing and iterating upon existing topological neural networks by sweeping over different base neural networks.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 76.58228% with 74 lines in your changes missing coverage. Please review.

Project coverage is 79.59%. Comparing base (04b085f) to head (4960607).

Files with missing lines	Patch %	Lines
...hmarkx/nn/backbones/combinatorial/gccn_onehasse.py	73.80%	33 Missing ⚠️
topobenchmarkx/nn/backbones/graph/identity_gnn.py	41.17%	20 Missing ⚠️
topobenchmarkx/nn/backbones/combinatorial/gccn.py	88.40%	16 Missing ⚠️
...nchmarkx/nn/wrappers/combinatorial/tune_wrapper.py	37.50%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #88      +/-   ##
==========================================
+ Coverage   77.45%   79.59%   +2.14%     
==========================================
  Files          98      105       +7     
  Lines        2404     2720     +316     
==========================================
+ Hits         1862     2165     +303     
- Misses        542      555      +13     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[gbg141]

Thank you!