Open lilleswing opened 4 years ago
Here's a couple of my observations so far:
We should make sure that there's a stable mechanism for splitting datasets that allows for easy benchmarking. This repo has some code that improves the stability (which was an issue in the original MoleculeNet):
The Moleculenet publication has accomplished much in terms of having standardized problems for supervised learning over chemical structures. However over the past couple of years we have seen some barriers to entry in using the datasets. How can we make it easier?
This issue can be a brainstorming page for how to make the MoleculeNet datasets more accessible to Machine Learning Practitioners.