How can I use per_species_rescale_shifts with exact values?

[pobo95]

Thank you for your good code!

These days, I have a problem of becoming gas on the melting point in NPT ensmeble.

And I'm thinking about per_species_rescale_shifts to solve this problem.

dataset_per_atom_total_energy_mean the default option does not seem to be a physical value.
Thats why I want to use exact values.

How can I use per_species_rescale_shifts with exact values?

When I use

global_rescale_shift: null
global_rescale_shift_trainable: false
global_rescale_scale: dataset_forces_rms
global_rescale_scale_trainable: false
per_species_rescale_shifts_trainable: false
per_species_rescale_scales_trainable: true
per_species_rescale_shifts: [Ge:-0.02859140,Sb:-0.03305661,Te:-0.03357790]
per_species_rescale_scales: dataset_per_species_forces_rms

It showed this KeyError: 'per_species_rescale_arguments_in_dataset_units' massage.

And when I used dataset_per_species_total_energy_mean option, it doen't work with lammps
Do you have any suggestions?

Thank you for your help.

[Linux-cpp-lisp]

Hi @pobo95 ,

Thanks for your interest in our code!

You need to set per_species_rescale_arguments_in_dataset_units: true to tell the code that the values you are providing for the per_species_rescale_shifts are in the same units of the dataset, rather than the scaled units of the model. Does it work if you set this value?

[pobo95]

Thank you for your quick answer @Linux-cpp-lisp!

No it doesn't work.

When I try

global_rescale_shift: null
global_rescale_shift_trainable: false
global_rescale_scale: dataset_forces_rms
global_rescale_scale_trainable: false
per_species_rescale_shifts_trainable: false
per_species_rescale_arguments_in_dataset_units: true
per_species_rescale_scales_trainable: true
per_species_rescale_shifts: [Ge: -0.02859140,Sb: -0.03305661,Te: -0.03357790]
per_species_rescale_scales: dataset_per_species_forces_rms

It got

Successfully loaded the data set of type ASEDataset(2000)...
Replace string dataset_per_species_forces_rms to tensor([0.4694, 0.6948, 0.6315], dtype=torch.float64)
Traceback (most recent call last):
File "/home/grad/pjpark/anaconda3/envs/nequip/bin/nequip-train", line 8, in
sys.exit(main())
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/scripts/train.py", line 81, in main
trainer = fresh_start(config)
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/scripts/train.py", line 209, in fresh_start
final_model = model_from_config(
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/model/_build.py", line 133, in model_from_config
model = builder(**params)
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/model/_scaling.py", line 219, in PerSpeciesRescale
shifts = torch.as_tensor(shifts)
RuntimeError: Could not infer dtype of dict

If I try

global_rescale_shift: null
global_rescale_shift_trainable: false
global_rescale_scale: dataset_forces_rms
global_rescale_scale_trainable: false
per_species_rescale_shifts_trainable: false
per_species_rescale_arguments_in_dataset_units: true
per_species_rescale_scales_trainable: true
per_species_rescale_shifts: [Ge:-0.02859140,Sb:-0.03305661,Te:-0.03357790]
per_species_rescale_scales: dataset_per_species_forces_rms

I got

Replace string dataset_per_species_forces_rms to tensor([0.4694, 0.6948, 0.6315], dtype=torch.float64)
Traceback (most recent call last):
File "/home/grad/pjpark/anaconda3/envs/nequip/bin/nequip-train", line 8, in
sys.exit(main())
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/scripts/train.py", line 81, in main
trainer = fresh_start(config)
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/scripts/train.py", line 209, in fresh_start
final_model = model_from_config(
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/model/_build.py", line 133, in model_from_config
model = builder(**params)
File "/home/grad/pjpark/anaconda3/envs/nequip/lib/python3.8/site-packages/nequip/model/_scaling.py", line 219, in PerSpeciesRescale
shifts = torch.as_tensor(shifts)
ValueError: too many dimensions 'str'

It got error when finding
Replace string dataset_per_energy_rms to tensor([somthine], dypte-torch.float64)
the number what I use [Ge: -0.02859140,Sb: -0.03305661,Te: -0.03357790] is eV unit

I'm using 0.5.3 v

Do you have any clue?

Thanks

[pobo95]

@Linux-cpp-lisp

Oh It works when I use
per_species_rescale_shifts: [-0.02859140,-0.03305661,-0.03357790]

Thank you for your help :)

[Linux-cpp-lisp]

Yes: the shifts should be in order of the assigned atom types, which depend on how you specified the chemical species.

[pobo95]

@Linux-cpp-lisp I totally understand thanks :)