Feat pt: Support multi structures input for single frame
Support multi structures input for one frame:
background: When we predict the properties of small molecules, sometimes multiple 3D conformations are generated from the SMILES format structure, but their targets are the same. In each epoch we would like to pick only one of the 3D configurations as input.
Impletation: When the file "multistru" is present in the data folder (like nopbc), "coord.npy" can be written in the format (nframe * struc_num * natom * 3), where struc_num is the number of 3D conformations for each structure.
Codecov Report
Attention: Patch coverage is 72.91667% with 13 lines in your changes are missing coverage. Please review.
Project coverage is 77.47%. Comparing base (
2851fb9) to head (fcce4df). Report is 1 commits behind head on devel.
| Files | Patch % | Lines |
|---|---|---|
| deepmd/utils/data.py | 76.19% | 10 Missing :warning: |
| deepmd/pt/utils/neighbor_stat.py | 50.00% | 3 Missing :warning: |
Additional details and impacted files
@@ Coverage Diff @@
## devel #3566 +/- ##
==========================================
- Coverage 77.49% 77.47% -0.02%
==========================================
Files 432 432
Lines 37164 37185 +21
Branches 1620 1620
==========================================
+ Hits 28801 28810 +9
- Misses 7495 7507 +12
Partials 868 868
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
@wanghan-iapcm Different 3D conformations of a molecule have exactly the same label, and each epoch takes one of all 3D conformations of this molecule. Because molecule may only has SMILES, the exact 3D sturcture is unknown.
@wanghan-iapcm Different 3D conformations of a molecule have exactly the same label, and each epoch takes one of all 3D conformations of this molecule. Because molecule may only has SMILES, the exact 3D sturcture is unknown.
I do not find any reason why different 3D configurations cannot be place in one batch.