Why lcurve.out have no information for testing like l2_e_tst, l2_e_tst in V2.0.0 ?

[zhao-w-en]

In V2.0.0 lcurve.out the headline is only # step rmse_trn rmse_e_trn rmse_f_trn lr , while V1.3.3 has # batch l2_tst l2_trn l2_e_tst l2_e_trn l2_e_tst l2_f_trn lr

[njzjz]

In v2.0 you need to set validation_data manually.

[zhao-w-en]

Thanks for your reply!
I remember that the last set of train system is used for testing in V1.3.3, so does that mean validation_data is corresponding to that set ? numb_test is number of frames used for the test during training. What is the relationship between batch_size of validation_data part and numb_test ?

[njzjz]

does that mean validation_data is corresponding to that set ?

Yes.

[njzjz]

What is the relationship between batch_size of validation_data part and numb_test

Good question. I think numb_test has been deprecated? cc @wanghan-iapcm

[wanghan-iapcm]

In v2.x the numb_test does not have any effect.

[zhao-w-en]

Oh, thank you! By the way, I also have some questions on numb_btch and batch_size of validation part. If numb_bath determine how many systems will be used for validation, batch_size determine what?
PS. maybe you could remark the parameter that it only suit for V1.3.3 from https://docs.deepmodeling.org/projects/deepmd/en/latest/train/train-input.html.