search

deepmodeling / dpdata

Manipulating multiple atomic simulation data formats, including DeePMD-kit, VASP, LAMMPS, ABACUS, etc.
https://docs.deepmodeling.com/projects/dpdata/
GNU Lesser General Public License v3.0
189 stars 128 forks source link

dp data cannot translate OUTCAR results in machine learning AIMD(MLAIMD) #586

Closed Suyoung999999999 closed 6 months ago

Suyoung999999999 commented 7 months ago

Summary

How to solve machine learning AIMD data into dpmd training data?

Detailed Description

Using machine learning AIMD for calculation will speed up the data preparation in the early stage of dpmd. However, I found that all frames could not be identified when using dpdata for format conversion. For example, I conducted an MLAIMD with 30,000 steps, but dpdata could only identify 269 of them. These 269 data were run without machine learning. So please ask how to use python script to achieve this feature, can you provide a python script?

Further Information, Files, and Links

No response

wanghan-iapcm commented 7 months ago

Could you please provide an example (the minimal one should be enough) output that may reproduce the issue? thanks!

Suyoung999999999 commented 7 months ago

Dear Professor Wang, Hello! The attachment below is my OUTCAR file using machine learning AIMD (MLAIMD). All data points cannot be converted using dp data, which may be the part that does not recognize machine learning. OUTCAR.zip 有关dp data如何实现ML-AIMD的数据转化-苏阳.pptx

Suyoung999999999 commented 7 months ago

Dear Professor Wang, Hello! I have uploaded an example on Github and look forward to your reply

su18812625523

@. | On 12/7/2023 09:14，Han @.> wrote：

Could you please provide an example (the minimal one should be enough) output that may reproduce the issue? thanks!

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

wanghan-iapcm commented 6 months ago

I find only 187 frames of your trajectory were labeled by dft, while other frames were labeled by ML. dpdata correctly load the 187 dft frames.

It does not make much to train against ML energy and forces. If you really need them you may modify this file .

Suyoung999999999 commented 6 months ago

Thanks for your reply

su18812625523

@. | On 12/13/2023 14:18，Han @.> wrote：

I find only 187 frames of your trajectory were labeled by dft, while other frames were labeled by ML. dpdata correctly load the 187 dft frames.

It does not make much to train against ML energy and forces. If you really need them you may modify this file .

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

njzjz commented 6 months ago

It has been supported by #282.