multiprocessing.pool error when trying to perform single point calculation with MLP

[gtusha]

Hello!
I am experiencing a technical issue when trying to perform a single point calculation on a structure with the MLP that I have trained previously.

The logfile shows no errors up to this point:
2023-12-01 16:20:37 md185 mlptrain.log[1398248] INFO Loaded calculator successfully

But then I get the following error:

File "/home/gtusha/softwares/mlp-train-main-final/mlp-train-main/mlptrain/configurations/configuration_set.py", line 414, in single_point
return self._run_parallel_method(function=_single_point_eval,
File "/home/gtusha/softwares/mlp-train-main-final/mlp-train-main/mlptrain/configurations/configuration_set.py", line 622, in _run_parallel_method
self[i] = result.get(timeout=None)
File "/home/gtusha/miniconda3/envs/mlptrain_final/lib/python3.9/multiprocessing/pool.py", line 771, in get
raise self._value
multiprocessing.pool.MaybeEncodingError: Error sending result: '<multiprocessing.pool.ExceptionWithTraceback object at 0x1527ee90b310>'. Reason: 'TypeError("cannot pickle 'PyCall.jlwrap' object")'

• The mlp is trained with ACE
• The error persists independently on the number of cores I use (1 or 2).
• The script works fine when I perform the single point with ORCA
• I am using julia 1.9.2
• I have tried it on different HPC systems

output.log

Thank you for the support!

[juraskov]

Hi Gers, thanks for reporting this. Did you use the same version of Julia for training the potential as you now use for the prediction? If you send me your input files I can also try to reproduce your error and give you more info.

Thanks!

Veronika

[gtusha]

Hi Veronika.
Yes, I have used the same Julia version in both computations.

I am enclosing the input file and the potential file.
files_sp.zip

Cheers

[juraskov]

The error is a result of a missing box in the ACE computations. It can be fixed by adding the following lines in the input, after the load_xyz function:

for structure in configurations:
structure.box = Box([100, 100, 100])

I will add the check of the box to the code.

Veronika

[gtusha]

Thanks a lot !