No change when restarting OTF with different training parameters

[usamaS97]

Hi MIR team,

I am using FLARE==0.2.4 with VASP 5.4.4 to train force field using ASE OTF.
Python 3.9.2 with GCC 8.3.0

During the initial run, the job timed out on the HPC due to large number of DFT Calls.
I am trying to restart otf training with a different std_tolerance_factor but it seems to have no effect.

I am writing the following in my Python Script:

from flare.ase.otf import ASE_OTF

test_otf= ASE_OTF.from_checkpoint('otf_checkpt.json')
test_otf.std_tolerance_factor= 2.5
test_otf.run()

In the 'otf.out', it still says the following:

Uncertainty tolerance: 1.5 times noise hyperparameter
Timestep (ps): 0.001
Number of frames: 2500
Number of atoms: 6
System species: {'O', 'Ti'}
Periodic cell (A):
[[4.598 0. 0. ]
[0. 4.598 0. ]
[0. 0. 2.96 ]]
Restart: 99

Any help would be appreciated

[Yu Xie]

Hi @usamaS97 ,

you can do

test_otf.std_tolerance = 2.5

instead of

test_otf.std_tolerance_factor= 2.5

There is a small confusion in the variable name and the input arg name

[usamaS97]

Hi @YuuuXie

Thank you so much!

It worked!