RiemannGFM

Get Started

First, install all the required pakages using

pip install -r requirements.txt

You need to Pretrain model before you start downstream tasks.

source ./scripts/pretrain.sh

The model you trained will be saved to directory ./checkpoints/.

If you want to define a custom name of saved model file, you can change the param --pretrained_model_path .

If you want to pretrain on other datasets, you can change the param --pretrain_datasets.

For the downstream tasks, if you want to run node classification, for example, you can use

source ./scripts/NC/github.sh

For Link Prediction task, you can run

source ./scripts/LP/citeseer.sh

If you do not want to load pretrained model for downstream tasks, you can add --load to bash script.

If you do not want to fine-tune pretrained model weights during training on downstream tasks, you can add --finetune to bash script.

Model Architecture

Experimental Results