pt: fix loss training when no data available

[iProzd]

Fix #3482 and #3483.

[codecov]

Codecov Report

Attention: Patch coverage is 73.58491% with 14 lines in your changes are missing coverage. Please review.

Project coverage is 77.49%. Comparing base (145f501) to head (ee81f4b).

Files	Patch %	Lines
deepmd/pt/loss/ener.py	63.15%	7 Missing ⚠️
deepmd/pt/loss/ener_spin.py	63.15%	7 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##            devel    #3571   +/-   ##
=======================================
  Coverage   77.48%   77.49%           
=======================================
  Files         432      432           
  Lines       37182    37202   +20     
  Branches     1620     1620           
=======================================
+ Hits        28812    28831   +19     
- Misses       7502     7503    +1     
  Partials      868      868           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[njzjz]

I understand it could fix #3483, but what was the problem with #3482?

[njzjz]

See @wanghan-iapcm's comment in #3047

[iProzd]

I understand it could fix #3483, but what was the problem with #3482?

Strangely, I can not reproduce #3482 even on current devel branch.

[iProzd]

See @wanghan-iapcm's comment in #3047

Fixed in d5e9f88

[njzjz]

I understand it could fix #3483, but what was the problem with #3482?

Strangely, I can not reproduce #3482 even on current devel branch.

I rechecked it and I think you are right. I may made a mistake here... I'll close #3482 directly.

[wanghan-iapcm]

How do you test if the loss work property when the label is absent?

[iProzd]

How do you test if the loss work property when the label is absent?

fix in 602a060