TypeError: __class__ assignment: 'FLARE_Atoms' object layout differs from 'Atom'

[Lenard Carroll]

Hi there, I am trying to use FLARE for Training a Gaussian Process from an AIMD Run. Here is an example of what I am trying to do so far https://colab.research.google.com/drive/16OBMTFhnPJBR54Ge9HBOA9KYUK9rYi9e?usp=sharing

I have read in my test .xyz file which consists of 3 frames, but I ultimately get the error of:
"TypeError: class assignment: 'FLARE_Atoms' object layout differs from 'Atom'"

Not sure how to fix this or how to go forward with it. Some help would be appreciated.

[Anders Johansson]

Hi Lenard,

I think ase.io.read will only read the last frame by default, so you need to give it index=":". Your code reads only one frame and tries to iterate over the atoms in it, rather than iterate over the three frames in the trajectory.

Btw: We are in the process of integrating the many-body version of FLARE, which you may want to check out on the development branch ;-) While the old 2+3-body code will remain available, we generally find that the new code is a direct upgrade (faster training, better accuracy, similar or better MD speed).

[Lenard Carroll]

Hi Anders, I am following along with the tutorial from https://mir-group.github.io/flare/tutorials/gpfa.html to get more acquainted with the code, but I have run into another issue. I got past the first issue which you helped me solve thanks, but now when I use: from flare.gp_from_aimd import TrajectoryTrainer TT = TrajectoryTrainer(frames=trajectory, gp = gp, rel_std_tolerance = 3, abs_std_tolerance=0, pre_train_on_skips=1) TT.run() print("Done!") AttributeError: 'FLARE_Atoms' object has no attribute 'indices_of_specie' I would really like to use FLARE in my work.

…

On Mon, Oct 24, 2022 at 2:36 AM Anders Johansson ***@***.***> wrote: Hi Lenard, I think ase.io.read will only read the last frame by default, so you need to give it index=":". Your code reads only one frame and tries to iterate over the atoms in it, rather than iterate over the three frames in the trajectory. Btw: We are in the process of integrating the many-body version of FLARE, which you may want to check out on the development branch ;-) While the old 2+3-body code will remain available, we generally find that the new code is a direct upgrade (faster training, better accuracy, similar or better MD speed). — Reply to this email directly, view it on GitHub <#325 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AUXMSSVC3YP5ZDUCWZ6MTKLWEXKZVANCNFSM6AAAAAARMM4IGA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[Yu Xie]

Hi @lenardcarroll, the issue is probably because in the development branch we removed the Structure class, which is used by TrajectoryTrainer and has the attribute indices_of_specie. To test the training from AIMD trajectory from our latest code, you can take a look at the offline training tutorial.