Does AML read frames from N2P2 datasets?

[lizhiqiang100]

Hi, developers,
Thank you for your excellent package. I am learning it, and I wonder that AML could read frames from N2P2 datasets, not just CP2K trajectories. I found the n2p2.py, runner.py files in the AML root directory. Can these files read the corresponding files like cp2k.py? Thanks in advance for your help.

Sincerely,
Zhiqiang

[cschran]

Dear Zhiqiang,

Thanks for your interest in our code. You can read in cp2k data with aml. The example from-trajectroy-to-model actually contains all relevant code:

fn_positions = dir_trj / 'zundel-pos-1.xyz'
fn_forces = dir_trj / 'zundel-frc-1.xyz'

frames = aml.read_frames_cp2k(fn_positions=fn_positions, fn_forces=fn_forces)
structures = aml.Structures.from_frames(frames, stride=stride_trj, probability=1.0)

Note that you might have to also add cell information (in atomic units) to read_frames_cp2k, if you deal with perodic systems.

All the best,
Christoph

[lizhiqiang100]

Thanks for your reply. My inital idea is that if I have a training set of N2P2, and I want to use aml to screen those representative structures😀. I am not sure this idea is reasonable and reliable for aml. What do you think of this idea?

[cschran]

if you have existing training data that you want to prune, you can simply read it in via

structures = aml.Structures.from_file('input.data')

and use that for the QbC process. Just make sure that you're setting n_candidate to something that is smaller than the training set size.

[lizhiqiang100]

Great thanks for your instruction， I can't wait to try it😄.