All configurations fail in the first cycle, and max_devi_f reaches the order of 10 to the first power

[12jscvb]

After the end of the first cycle of dpgen, all configuration failures occurred, and the reason could not be found. A screenshot of the error and the param.json file are attached below. Thank you for help!

[njzjz]

The max_devi_f is too large in the first step, which means the training data or the initial structure is completely wrong.

[12jscvb]

I ran the initial dataset through cp2kaimd with a time step of 1fs and ran 200 steps，Is there an error in this step?

[njzjz]

You could first check if dp test reports the expected RMSE.
https://docs.deepmodeling.com/projects/deepmd/en/master/test/test.html

[12jscvb]

I did not find the expected RMSE in the dp test. What should I do? Additionally, I use relative model error in param.json, f_ Low and f_  hi was 0.4 and 0.8, it was found that thirty configurations would be candidates, while the rest were failed configurations

…

---Original--- From: "Jinzhe ***@***.***> Date: Sat, Oct 28, 2023 05:02 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [deepmodeling/dpgen] All configurations fail in the first cycle,and max_devi_f reaches the order of 10 to the first power (Issue #1363) You could first check if dp test reports the expected RMSE. https://docs.deepmodeling.com/projects/deepmd/en/master/test/test.html — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>