extxyz format and without force data

[esma-dd]

Hi, thanks for efforts for nequip

I need to run nequip on a data set of xyz coordinates and energy data, force data is not available. I tried removed forces:R:3 and force columns from extxyz data file, it did not work, it gave no forces is found.

Piece of data set and failed run output are stated below.

21
Properties=species:S:1:pos:R:3 energy=-406757.5912640221
C 2.2393 -0.3791 0.263
C 0.8424 1.9231 -0.4249
C 2.8709 0.8456 0.2722
C 2.1751 1.9935 -0.0703
C -3.4838 0.4953 -0.0896
C 0.891 -0.4647 -0.0939
C 0.1908 0.6991 -0.4402
O -0.9633 -1.8425 -0.4185
O -1.6531 0.8889 1.3406
O 0.8857 -2.8883 0.2267
C 0.209 -1.772 -0.1069
C -2.0185 0.6853 0.2071
O -1.1189 0.6285 -0.7886
H 0.3962 -3.7219 0.2035
H 2.7867 -1.2719 0.5268
H 0.3069 2.8224 -0.6911
H 3.913 0.9108 0.5482
H 2.6781 2.9492 -0.0604
H -3.736 -0.5623 -0.012
H -4.0763 1.0637 0.6273
H -3.6988 0.8471 -1.0986
21
Properties=species:S:1:pos:R:3 energy=-406757.03429293
C 2.2407929 -0.3792177 0.26220864
C 0.84189833 1.92928051 -0.42386094
C 2.87186381 0.84557865 0.27150796
C 2.17214107 1.99676402 -0.06799014
C -3.48395765 0.49426137 -0.09081787
C 0.89348535 -0.46899634 -0.09478387
C 0.19230078 0.69800944 -0.43901899
O -0.9630015 -1.84111352 -0.41665285
O -1.65334837 0.8862533 1.34035521
O 0.8841712 -2.88668415 0.22312671
C 0.2123762 -1.77410418 -0.10486024
C -2.02150244 0.68643118 0.21080257
O -1.11879805 0.62737142 -0.79006091
H 0.38546463 -3.73362241 0.23042788
H 2.77228461 -1.28617246 0.52334609
H 0.31047555 2.80672929 -0.70986827
H 3.9204822 0.91580973 0.55342684
H 2.67876592 2.93630784 -0.08058516
H -3.74977798 -0.56203982 -0.01575305
H -4.07244665 1.09156644 0.60937952
H -3.69170686 0.85806722 -1.09188342

Successfully loaded the data set of type ASEDataset(10)...
Replace string dataset_per_atom_total_energy_mean to -19369.17009832072
Atomic outputs are scaled by: [H, C, O: None], shifted by [H, C, O: -19369.170098].
Traceback (most recent call last):
  File "/usr/local/bin/nequip-train", line 8, in <module>
    sys.exit(main())
  File "/usr/local/lib/python3.10/dist-packages/nequip/scripts/train.py", line 94, in main
    trainer = fresh_start(config)
  File "/usr/local/lib/python3.10/dist-packages/nequip/scripts/train.py", line 222, in fresh_start
    final_model = model_from_config(
  File "/usr/local/lib/python3.10/dist-packages/nequip/model/_build.py", line 133, in model_from_config
    model = builder(**params)
  File "/usr/local/lib/python3.10/dist-packages/nequip/model/_scaling.py", line 20, in RescaleEnergyEtc
    return GlobalRescale(
  File "/usr/local/lib/python3.10/dist-packages/nequip/model/_scaling.py", line 80, in GlobalRescale
    computed_stats = _compute_stats(
  File "/usr/local/lib/python3.10/dist-packages/nequip/model/_scaling.py", line 330, in _compute_stats
    values = dataset.statistics(
  File "/usr/local/lib/python3.10/dist-packages/nequip/data/_dataset/_base_datasets.py", line 413, in statistics
    raise RuntimeError(
RuntimeError: The field key `forces` is not present in this dataset

[cw-tan]

Hi @esma-dd

Thank you for your interest in our code!

Do you have global_rescale_scale: dataset_forces_rms included in your config? This will cause nequip to compute the force RMS of the dataset to use it as a global energy scale. If you don't have forces in the dataset, but this argument is provided, it could be a reason for the error. I'm also just guessing -- it would be easier to diagnose if we could see your config file.

[esma-dd]

Hi @cw-tan, thank you so much for such a prompt reply. "global_rescale..." solved the issue.

[cw-tan]

Great! You may want to look at the description of the behavior here

nequip/configs/full.yaml

Line 361 in 1e150cd

# If not provided, defaults to either dataset_force_rms or dataset_total_energy_std, depending on whether forces are being trained.

Looks like it'll default to the total_energy's standard deviation if you leave that flag empty.

[esma-dd]

Exactly where I modified, thank you again!