Please select To the mobile version | Continue to access the desktop computer version

BBS-DeePMD

 Forgot password?
 Register
Search
View: 75|Reply: 2
Collapse the left

deepmd force error not decline

[Copy link]
Post time: 2020-07-29 10:36:18
| Show all posts |Read mode
Hi!
I trained my model with deepmd-kit. The parameter is set as follow. But as the lcuve.out, the force error of both training data and test data does not decline after just 3000 batch. And the training error is still about 0.2ev/A while the the standard deviations of my force data is about 1ev/A. I have tried to change the stop batch and learning rate decay together or to use some  similar data(133 to 532 atoms, 6000 to 16000 frames), but the training error is still above 0.2ev/A.
Could you please help me  see what's wrong with my training paremeter and how to decrease the force error?
Thank you.

{
    "_comment": " model parameters",
    "model": {
        "type_map":        ["Al", "Cl","C"],
        "descriptor" :{
            "type":                "se_a",
            "sel":                [120, 120,120],
            "rcut_smth":        5.80,
            "rcut":                6.00,
            "neuron":                [25, 50, 100],
            "resnet_dt":        false,
            "axis_neuron":        16,
            "seed":                1,
            "_comment":                " that's all"
        },
        "fitting_net" : {
            "neuron":                [240, 240, 240],
            "resnet_dt":        true,
            "seed":                1,
            "_comment":                " that's all"
        },
        "_comment":        " that's all"
    },

    "learning_rate" :{
        "type":                "exp",
        "start_lr":        0.001,
        "decay_steps":        5000,
        "decay_rate":        0.95,
        "_comment":        "that's all"
    },

    "loss" :{
        "start_pref_e":        0.02,
        "limit_pref_e":        1,
        "start_pref_f":        1000,
        "limit_pref_f":        1,
        "start_pref_v":        0,
        "limit_pref_v":        0,
        "_comment":        " that's all"
    },

    "_comment": " traing controls",
    "training" : {
        "systems":        ["../Al1C512Cl4/","../Al3C512Cl12/","../Al4C512Cl16/"],
        "set_prefix":        "set",   
        "stop_batch":        1000000,
        "batch_size":        "auto",

        "seed":                1,

        "_comment": " display and restart",
        "_comment": " frequencies counted in batch",
        "disp_file":        "lcurve.out",
        "disp_freq":        100,
        "numb_test":        10,
        "save_freq":        1000,
        "save_ckpt":        "model.ckpt",
        "load_ckpt":        "model.ckpt",
        "disp_training":true,
        "time_training":true,
        "profiling":        false,
        "profiling_file":"timeline.json",
        "_comment":        "that's all"
    },

    "_comment":                "that's all"
}


forceerror.jpg
energyerror.jpg
Reply

Use magic Report

Post time: 2020-07-29 16:02:20
| Show all posts
How do you calculate yours standard deviations ? thanks
Reply Support Not support

Use magic Report

 Author| Post time: 2020-07-30 12:53:00
| Show all posts
keng 2020-07-29 16:02
How do you calculate yours standard deviations ? thanks

I put force.raw data to an excel shell.
Reply Support Not support

Use magic Report

You have to log in before you can reply Login | Register

Points Rules

Dark room|DeePMD

2020-08-12 00:48 GMT+8 , Processed in 0.111049 sec., 21 queries .

Quick Reply To Top Return to the list