search

divelab / DIG

A library for graph deep learning research
https://diveintographs.readthedocs.io/
GNU General Public License v3.0
1.82k stars 281 forks source link

ComENet: Molecule3D split #172

Closed fmocking closed 1 year ago

fmocking commented 1 year ago

Hello, great work on ComENet, thanks for sharing! I meant to ask in the poster session in NeurIPS, can you provide the splits for Molecule3D? In your poster it stated as 'random', i believe it would be great to have some canonical split (similar to QM9) for researchers to use. I believe ComENet is great start for this. Also similar to #155 it would be great to have the pipeline ready for researchers to test.

Best regards,

limei0307 commented 1 year ago

Hi @fmocking,

Thanks for your interest in our work! Molecule3D is included in another package by our lab here. We may adopt the exact dataset we used in ComENet in this package later.

Thanks

fmocking commented 1 year ago

Hi @limei0307 thank you for your prompt response. After downloading the raw data for Molecule3D I see the indices are available to use. Thanks!