This repository includes changes to Graphormer's and FraudGT's attention mechanisms to perform node-inductive tasks for Relational Deep Learning. This implementation is part of CSE3000 - Research Project
- Create a new Conda environment, and activate.
conda env create -f env.yml
conda activate relbench-env
- Install Pytorch and Pytorch Geometric
conda install pytorch==2.2.2 torchvision==0.17.2 torchaudio==2.2.2 pytorch-cuda=11.8 -c pytorch -c nvidia
conda install pyg -c pyg
pip install -r requirements.txt
pip install pytorch-frame sentence_transformers relbenchCreate a venv, then run
source venv/bin/activateThen run the following
pip install torch==2.6.0 torchvision torchaudio torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.6.0+cpu.html
pip install pytorch-frame sentence_transformers relbench wandbpython3 main.py \
--model <local|global> \
--dataset <f1|…> \
--task <driver-position|…> \
--save_artifacts \
--num_workers <int> \
--eval_freq <int> \
--lr <float> \
--epochs <int> \
--optimiser <adam|adamW> \
--batch_size <int> \
--channels <int> \
--aggr <sum|mean|max|…> \
--num_layers <int> \
--num_layers_pre_gt <int> \
--num_neighbors <n₁ n₂ …> \
--temporal_strategy uniform \
--rev_mp \
--port_numbering \
--ego_ids \
--edge_features \
--dropouts <d_local d_global d_attn> \
--head <HeteroGNNNodeHead|HeteroGNNNodeRegressionHead> \
--early_stopping \
--seed <int>For example
python3 main.py \
--model graphormer \
--save_artifacts \
--num_workers 12 \
--dataset avito \
--task ad-ctr \
--eval_freq 4 \
--batch_size 64 \
--epochs 15 \
--channels 64 \
--num_neighbors 50 50 \
--rev_mp \
--edge_features \
--dropouts 0.1 0.1 0.1 \
--head HeteroGNNNodeRegressionHead \
--seed 1For all scripts, look at /scripts