mcneela / mixed-curvature-gcn

Graph Convolutional Networks on Cartesian products of constant curvature manifolds.

Mixed-Curvature Product Space GCN

Adapted from https://github.com/HazyResearch/hgcn and further forked from https://github.com/fal025/product_hgcn.

The dependencies are the following:

virtualenv -p [PATH to python3.7 binary] hgcn
source hgcn/bin/activate
pip install -r requirements.txt

To run additional tests, please use train.py This script trains models for link prediction and node classification tasks. Metrics are printed at the end of training or can be saved in a directory by adding the command line argument --save=1.

For our purposes, the following command may be more useful than others:

python3 train.py --task lp --dataset <dataset_name> --model HGCN --lr 0.01 --dim <num_dim> --num-layers 2 --act relu --bias 0 --dropout 0.5 --weight-decay 0.001 --manifold <choice_of_prod_space> --log-freq 5 --cuda 0 --c 1

To specify a choice of product space use a string of the form <H{x}E{y}S{z}> x, y, and z indicate the number of copies of Hyperbolic, Euclidean, and Spherical space, respectively. An example input would be H1E2S2.

Citation

The modified Mixed-Curvature Product Space GCN is described in the following preprint:
Product Manifold Representations for Learning on Biological Pathways
Daniel McNeela, Frederic Sala⁺, Anthony Gitter⁺.
arXiv:2401.15478. 2024

See also the hgcn repository references.

License

The code is available under the Apache License 2.0. Some of the source code is derived from the unlicensed hgcn and product_hgcn repositories. The named contributors of those repositories have been added to the license copyright.

The hgcn repository notes additional code was forked from the following repositories

• pygcn
• gae
• hyperbolic-image-embeddings
• pyGAT
• poincare-embeddings
• geoopt