Data Stating Duration for Larger Batch Sizes

[npiroozan]

With larger data sets, in training, data stating can take a very long time especially for larger batch sizes. Is there any way to alleviate this? Thank you very much for your time.

[wanghan-iapcm]

One may consider reducing the data_stat_nbatch
see https://docs.deepmodeling.com/projects/deepmd/en/latest/train/train-input.html

[npiroozan]

Thank you very much indeed. I would like to understand what is happening in this phase of the simulation. What happens in data_stating? Thank you very much.

[njzjz]

It does two things:
(1) Normalizes the environment matrices;
(2) Determines the initial atomic energy biases (trainable) for each atom type.