whole training dataset

[shijiale0609]

The training data path in the code configuration is the author's local path, not a public URL.

starling/configs/dataloader/dataloader.yaml:4 points to

/work/bnovak/projects/sequence2ensemble/lammps_data/combined_data/data

How to get the whole training dataset?

[alexholehouse]

Hi Jiale!

Very good question - so training/test/val is about 2 TBs total (after compression), so too big for the usual sources where we'd deposit open data. We are trying to figure out options to make this work, but it's not trivial!

We have all the data assembled, but want to double-check that the configs used are up to date. We are just about to travel to the Biophysical Society meeting, so I ask if you can wait a couple of weeks as we figure out (a) what the best options here are and (b) actually move the data to a public destination. Thanks for your patience! We'll follow up here asap.

[wengong-jin]

Thank you for your great work! We are also very interested in this dataset and wonder if it is possible to share at least the pdb / xtc / mdtraj files of your data soon. I think huggingface can host datasets of that size.