merge compute_output_stat

[njzjz]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 96.49123% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 75.77%. Comparing base (897fcc5) to head (b7dbc8a).

Files	Patch %	Lines
deepmd/pt/model/task/ener.py	88.88%	1 Missing ⚠️
deepmd/tf/fit/dos.py	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #3310      +/-   ##
==========================================
- Coverage   75.83%   75.77%   -0.06%     
==========================================
  Files         415      382      -33     
  Lines       34862    30121    -4741     
  Branches     1614      909     -705     
==========================================
- Hits        26437    22824    -3613     
+ Misses       7557     6769     -788     
+ Partials      868      528     -340     

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

[wanghan-iapcm]

I don't know how to compute for dipole and polar - TF has no such method.

I would propose compute the avg (like the atom_bias_e) and std of label. By far we remove the avg, it is also possible to scale the output of the fitting net by the std.

The polar of the TF uses the statistics of the diagonal of the avg.

[wanghan-iapcm]

I realize that the std for each type is not available. We may need to uniformly scale the atomic output based on statistics.

[wanghan-iapcm]

please check my comments on deepmd/utils/out_stat.py