This repository is for PCQM4Mv2 OGB-LSC @ NeurIPS 2022 | Open Graph Benchmark (stanford.edu).

Installation

The dependencies are created by miniconda.

python==3.8
ogb==1.3.4
pytorch==1.11.0
pyg==2.0.4
sympy==1.11.1
rdkit==2022.3.5

About Data

Data should be organized as follows,

--data
--pcqm4m-v2
    --raw
        data.csv.gz postion_all.pkl(generated by code)
    --processed_3d_position
        geometric_data_processed.pt(generated by code)
    RELEASE_v1.txt
    split_dict.pt

Quick Start

Make sure the data is organized as described above.

Training

Changing hyper-parameters and starting training in run.sh.

cd src
sh run.sh

Test inference

At inference time, we use rdkit to generate position data, it will take some time.

cd src
sh inference.sh