Adapted from https://github.com/HazyResearch/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
For now, the choice of product space should be the ratio between three spaces <E,S,P> corresponding to Euclidean, Spherical and Hyperboloid(PoincareBall) and the sum should be a divisor of your dimension input. An example input would be P1S1E2, where 4 divids 16.
The dataset name for now can be chosen from cora for nc and lp, pubmed for nc (or lp or if one have memory > 32G due to the large number of nodes in the dataset; otherwise the initialization step ), disease_lp for lp, disease_nc for nc, airport for lp and nc.
The result can be replicate through the running the run_test.py script.