Add QM9 datasets

[yuehhua]

Support Qm9 dataset

Ref: http://quantum-machine.org/datasets/ or https://pytorch-geometric.readthedocs.io/en/latest/modules/datasets.html

[CaiYitao]

do you mean we will be able to load the GEOM datasets like qm9.pkl iso17 or drug, in GraphMLDatasets?

[yuehhua]

Yes, you can list datasets you want GraphMLDatasets to support.
GraphMLDatasets will give graph, node features, etc., separately by GraphMLDatasets's API, e.g. graphdata(QM9()), node_feature(QM9()).
So, you can organize them in your own way, instead of bounded by Data class from torch geometric.

[CaiYitao]

At the moment I am dealing with .sdf and .mol kind of datasets which can be loaded by MolecularGraph.jl, but the GEOM datasets(including qm9.pkl, iso17.pkl, drug.pkl, these datsets are .pkl file) from that ConfGF package can not be loaded.

[yuehhua]

So, you need qm9, iso17, and drug?

[CaiYitao]

yeah, it would be great if we have those, the original dataset GEOM from Harvard, it contains qm9 and drug.
after you add them, can we open those qm9.pkl, iso17.pkl file in Julia by Pickle.jl or by GraphMLDatasets.jl?
Thank you!

[yuehhua]

It's a bite tricky to support the processed version. Does the original dataset work in your condition?

[CaiYitao]

I have not tried original data yet, Maybe I will try to use .sdf dataset only.
Thank you!

[CaiYitao]

I noticed that you also have GeometricFlux, have you considered using Optimal Transport to connect Geometric deep learning with Neural ODE? I found the course of optimal transport is very inspiring, so would like to share with you : )