[Feature Request] support FeNNol

[njzjz]

Summary

Support FeNNol (convert from dpdata LabeledSystem to this format) per requested by the reviewer. Unit tests are encouraged.

Detailed Description

Below is FeNNol's documentation.

rmd17_aspirin_01.pkl.zip

Dataset structure The file `rmd17_aspirin_01.pkl' contains a split of the aspirin dataset from the revMD17 dataset. This split contains 1000 structures for training and 50 structures for validation. Energy units are in kcal/mol and distances in Angstroms.

This dataset is formatted to work with FeNNol's training system. You can inspect it with the following Python code:

import pickle

with open('rmd17_aspirin_01.pkl', 'rb') as f:
    data = pickle.load(f)

print(data.keys()) # dict_keys(['training', 'validation','description'])
print(len(data['training'])) # 1000
print(len(data['validation'])) # 50

print(data['training'][0].keys()) # dict_keys(['species', 'coordinates', 'formation_energy', 'shifted_energy', 'forces'])

When preparing your own dataset, make sure to use a similar formatting.

Further Information, Files, and Links

No response