search

deepmodeling / deepmd-kit

A deep learning package for many-body potential energy representation and molecular dynamics
https://docs.deepmodeling.com/projects/deepmd/
GNU Lesser General Public License v3.0
1.41k stars 486 forks source link

[feature request] Usage of `sel` when restarting from previous frozen model #3939

Open wanghan-iapcm opened 2 days ago

wanghan-iapcm commented 2 days ago

Discussed in https://github.com/deepmodeling/deepmd-kit/discussions/3935

Originally posted by **francescomambretti** July 1, 2024 Dear all, I have a question concerning the usage of `sel` parameter. Let's say that I have a model 1, as a result of training on a given dataset. This model has a given value for `sel`. Let us suppose that I want to train a model 2, starting from the frozen model 1 using the keyword `--init-frz-model`, where the dataset now contains a new set of configurations which require a different (larger) value of `sel`, at least for some chemical elements of the system. Is it possible to use a new value for `sel` while restarting from model 1, or do I have to train from scratch on the new dataset (containing all the configurations) using the larger `sel`? Thanks!
njzjz commented 2 days ago

I think it has been supported in #1573 with some restraints.

wanghan-iapcm commented 1 day ago

I agree. Unfortunately, we do not have a good design for this feature at the descriptor interface, thus it seems that only se_a support changing the sel.