provide an option to skip neighbor stat

[njzjz]

Resolves #1088.

Also fixes a bug in update_sel.

[codecov-commenter]

Codecov Report

Merging #1313 (2111e46) into devel (843a3c5) will decrease coverage by 11.33%.
The diff coverage is n/a.

❗ Current head 2111e46 differs from pull request most recent head c8ded77. Consider uploading reports for the commit c8ded77 to get more accurate results

@@             Coverage Diff             @@
##            devel    #1313       +/-   ##
===========================================
- Coverage   75.62%   64.28%   -11.34%     
===========================================
  Files          91        5       -86     
  Lines        7494       14     -7480     
===========================================
- Hits         5667        9     -5658     
+ Misses       1827        5     -1822     

Impacted Files	Coverage Δ
deepmd/entrypoints/main.py
deepmd/entrypoints/train.py
deepmd/infer/deep_tensor.py
deepmd/entrypoints/__init__.py
deepmd/cluster/__init__.py
deepmd/descriptor/se_a.py
deepmd/fit/wfc.py
source/op/_prod_virial_se_a_grad.py
deepmd/entrypoints/test.py
deepmd/op/__init__.py
... and 76 more

---

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 843a3c5...c8ded77. Read the comment docs.

[denghuilu]

Do we need to support the model compression workflow after skipping neighbor stat? For example, using training script as well as the training data to recompute the neighbor stat.

[njzjz]

Do we need to support the model compression workflow after skipping neighbor stat? For example, using training script as well as the training data to recompute the neighbor stat.

We have already supported it...

[njzjz]

Oh, currently one needs to provide the script even if it has been stored

[denghuilu]

Do we need to support the model compression workflow after skipping neighbor stat? For example, using training script as well as the training data to recompute the neighbor stat.

We have already supported it...

If there's no neighbor stat in the original training process, the model compression will not be abled due to the lack of min_nbor_stat. We need to recalculate min_nbor_stat in this situation.

The calculation of nborstat is necessary, either during training or during model compression.

[njzjz]

Do we need to support the model compression workflow after skipping neighbor stat? For example, using training script as well as the training data to recompute the neighbor stat.

We have already supported it...

If there's no neighbor stat in the original training process, the model compression will not be abled due to the lack of min_nbor_stat. We need to recalculate min_nbor_stat in this situation.

The calculation of nborstat is necessary, either during training or during model compression.

deepmd-kit/deepmd/entrypoints/compress.py

Line 88 in c8ded77

t_min_nbor_dist = get_min_nbor_dist(jdata, get_rcut(jdata))

Isn't it?

[denghuilu]

Do we need to support the model compression workflow after skipping neighbor stat? For example, using training script as well as the training data to recompute the neighbor stat.

We have already supported it...

If there's no neighbor stat in the original training process, the model compression will not be abled due to the lack of min_nbor_stat. We need to recalculate min_nbor_stat in this situation.
The calculation of nborstat is necessary, either during training or during model compression.

deepmd-kit/deepmd/entrypoints/compress.py

Line 88 in c8ded77

t_min_nbor_dist = get_min_nbor_dist(jdata, get_rcut(jdata))

Isn't it?

Do we need to support the model compression workflow after skipping neighbor stat? For example, using training script as well as the training data to recompute the neighbor stat.

We have already supported it...

If there's no neighbor stat in the original training process, the model compression will not be abled due to the lack of min_nbor_stat. We need to recalculate min_nbor_stat in this situation.
The calculation of nborstat is necessary, either during training or during model compression.

deepmd-kit/deepmd/entrypoints/compress.py

Line 88 in c8ded77

t_min_nbor_dist = get_min_nbor_dist(jdata, get_rcut(jdata))

Isn't it?

yes, no problem now