Skip to content

add atom energy bias to type embedding energy#1592

Merged
wanghan-iapcm merged 3 commits intodeepmodeling:develfrom
njzjz:type-embd-bias
Mar 25, 2022
Merged

add atom energy bias to type embedding energy#1592
wanghan-iapcm merged 3 commits intodeepmodeling:develfrom
njzjz:type-embd-bias

Conversation

@njzjz
Copy link
Member

@njzjz njzjz commented Mar 24, 2022

Fix #684, where systems have different atom_numb. After this fix, RMSE of energy should be quickly decreased in the very beginning.

Fix deepmodeling#684, where systems have different `atom_numb`. After this fix, RMSE should be quickly decreased in the very beginning.
@codecov-commenter
Copy link

codecov-commenter commented Mar 24, 2022

Codecov Report

Merging #1592 (c91bad6) into devel (c49533e) will decrease coverage by 0.00%.
The diff coverage is 80.00%.

@@            Coverage Diff             @@
##            devel    #1592      +/-   ##
==========================================
- Coverage   76.75%   76.75%   -0.01%     
==========================================
  Files          93       93              
  Lines        7790     7794       +4     
==========================================
+ Hits         5979     5982       +3     
- Misses       1811     1812       +1     
Impacted Files Coverage Δ
deepmd/fit/ener.py 92.24% <80.00%> (-0.31%) ⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c49533e...c91bad6. Read the comment docs.

@wanghan-iapcm wanghan-iapcm requested a review from iProzd March 25, 2022 01:32
@wanghan-iapcm wanghan-iapcm merged commit c17873d into deepmodeling:devel Mar 25, 2022
@njzjz njzjz deleted the type-embd-bias branch March 25, 2022 04:57
njzjz added a commit to njzjz/deepmd-kit that referenced this pull request Mar 28, 2022
Following deepmodeling#1592, I noticed that constants cannot be recovered when restart training.
amcadmus pushed a commit that referenced this pull request Mar 29, 2022
* convert constants (which may change) to variables

Following #1592, I noticed that constants cannot be recovered when restart training.

* add dtype to variables

* cast int32

* bugfix

* init atom_ener
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[BUG] Loss of energy and force failed to decline with type-embedding

3 participants