Work as part of ANL summer 2022 research with emphasis on utilizing GraphGym + DeepHyper to perform HPO on graph neural networks.
The emphasis of the project is on Graph representation learning using Transformers. Leveraging the work from Recipe for a General, Powerful, Scalable Graph Transformer, this project attempts to utilize the Positional Encoders, and Structural Encoders provided in GraphGPS and benchmark various molecular property prediction datasets. The contribution of this project is to provide an integration of GraphGPS with DeepHyper, in order to have a GNN pipelines ready for Hyperparamter Optimization.
PyTorchPyTorch-GeometricDeepHyper
# Install Pytorch
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116
# Install Pytorch Geometric
pip install -q torch-scatter -f https://data.pyg.org/whl/torch-${TORCH}.html
pip install -q torch-sparse -f https://data.pyg.org/whl/torch-${TORCH}.html
pip install -q git+https://github.com/pyg-team/pytorch_geometric.git
# Install DeepHyper
conda install gxx_linux-64 gcc_linux-64
pip install pip --upgrade
pip install -q deephyper["analytics"]
# Install rdkit for the datasets
pip install -q rdkit-pypiMoleculetNet: MoleculeNet is a benchmark specially designed for testing machine learning methods of molecular properties. As we aim to facilitate the development of molecular machine learning method, this work curates a number of dataset collections, creates a suite of software that implements many known featurizations and previously proposed algorithms. All methods and datasets are integrated as parts of the open source DeepChem package(MIT license).
Reference: https://moleculenet.org/datasets-1
Resource: https://pytorch-geometric.readthedocs.io/en/latest/modules/datasets.html
TBA
- L. Rampášek, et. al, Recipe for a General, Powerful, Scalable Graph Transformer, ArXiv, arxiv:2205.12454, 2022.
- J. You, et. al, Design Space for Graph Neural Networks, ArXiv, arxiv:2011.08843, 2021.