deepmodeling
Large energy and force Errors when using Atom type embedding (test for deepmd-2.0beta)
>Summary
The energy and force Errors for training using Atom type embedding are large.
>Deepmd-kit version input.json lcurve.out
2.0.0.b0, conda
>input.json
{
"model": {
"type_map": [
"S"
],
"type_embedding":{
"neuron": [2, 4, 8],
"resnet_dt": false,
"seed": 1
},
"descriptor": {
"type": "se_a_tpe",
"sel": [
128
],
"rcut_smth": 0.5,
"rcut": 6.5,
"neuron": [
25,
50,
100
],
"resnet_dt": false,
"axis_neuron": 16,
"seed": 722586222
},
"fitting_net": {
"neuron": [
240,
240,
240,
240
],
"resnet_dt": true,
"seed": 711230366
}
},
"learning_rate": {
"type": "exp",
"start_lr": 0.001,
"decay_steps": 5000,
"stop_lr": 3.51e-8
},
"loss": {
"type":"ener",
"start_pref_e": 0.02,
"limit_pref_e": 1,
"start_pref_f": 1000,
"limit_pref_f": 1,
"start_pref_v": 0.0,
"limit_pref_v": 0.0
},
"training": {
"seed": 10,
"disp_file": "lcurve.out",
"stop_batch": 1000000,
"disp_freq": 1000,
"numb_test": 5,
"save_freq": 1000,
"save_ckpt": "model.ckpt",
"profiling": false,
"profiling_file": "timeline.json",
"training_data": {
"systems": ["../data.iters/iter.000000/02.fp/data.002",
"../data.iters/iter.000000/02.fp/data.032",
"../data.iters/iter.000000/02.fp/data.000",
"../data.iters/iter.000000/02.fp/data.034",
"../data.iters/iter.000001/02.fp/data.004",
"../data.iters/iter.000001/02.fp/data.036",
"../data.iters/iter.000001/02.fp/data.006",
"../data.iters/iter.000001/02.fp/data.038",
"../data.iters/iter.000002/02.fp/data.016",
"../data.iters/iter.000002/02.fp/data.048",
"../data.iters/iter.000002/02.fp/data.018",
"../data.iters/iter.000002/02.fp/data.050",
"../data.iters/iter.000003/02.fp/data.022",
"../data.iters/iter.000003/02.fp/data.054",
"../data.iters/iter.000003/02.fp/data.052",
"../data.iters/iter.000003/02.fp/data.020",
"../data.iters/iter.000004/02.fp/data.058",
"../data.iters/iter.000004/02.fp/data.024",
"../data.iters/iter.000004/02.fp/data.026",
"../data.iters/iter.000004/02.fp/data.056",
"../data.iters/iter.000005/02.fp/data.060",
"../data.iters/iter.000005/02.fp/data.028",
"../data.iters/iter.000005/02.fp/data.030",
"../data.iters/iter.000005/02.fp/data.062",
"../data.iters/iter.000006/02.fp/data.010",
"../data.iters/iter.000006/02.fp/data.008",
"../data.iters/iter.000006/02.fp/data.042",
"../data.iters/iter.000006/02.fp/data.040",
"../data.iters/iter.000007/02.fp/data.044",
"../data.iters/iter.000007/02.fp/data.014",
"../data.iters/iter.000007/02.fp/data.046",
"../data.iters/iter.000007/02.fp/data.012"],
"batch_size": 1
},
"validation_data":{
"systems": ["../data.iters/iter.000000/02.fp/data.002",
"../data.iters/iter.000000/02.fp/data.032",
"../data.iters/iter.000000/02.fp/data.000",
"../data.iters/iter.000000/02.fp/data.034",
"../data.iters/iter.000001/02.fp/data.004",
"../data.iters/iter.000001/02.fp/data.036",
"../data.iters/iter.000001/02.fp/data.006",
"../data.iters/iter.000001/02.fp/data.038",
"../data.iters/iter.000002/02.fp/data.016",
"../data.iters/iter.000002/02.fp/data.048",
"../data.iters/iter.000002/02.fp/data.018",
"../data.iters/iter.000002/02.fp/data.050",
"../data.iters/iter.000003/02.fp/data.022",
"../data.iters/iter.000003/02.fp/data.054",
"../data.iters/iter.000003/02.fp/data.052",
"../data.iters/iter.000003/02.fp/data.020",
"../data.iters/iter.000004/02.fp/data.058",
"../data.iters/iter.000004/02.fp/data.024",
"../data.iters/iter.000004/02.fp/data.026",
"../data.iters/iter.000004/02.fp/data.056",
"../data.iters/iter.000005/02.fp/data.060",
"../data.iters/iter.000005/02.fp/data.028",
"../data.iters/iter.000005/02.fp/data.030",
"../data.iters/iter.000005/02.fp/data.062",
"../data.iters/iter.000006/02.fp/data.010",
"../data.iters/iter.000006/02.fp/data.008",
"../data.iters/iter.000006/02.fp/data.042",
"../data.iters/iter.000006/02.fp/data.040",
"../data.iters/iter.000007/02.fp/data.044",
"../data.iters/iter.000007/02.fp/data.014",
"../data.iters/iter.000007/02.fp/data.046",
"../data.iters/iter.000007/02.fp/data.012"],
"batch_size": 1,
"numb_btch": 3,
"_comment": "that's all"
},
"_comment": "that's all"
}
}
lcurve.out
step rmse_val rmse_trn rmse_e_val rmse_e_trn rmse_f_val rmse_f_trn lr 0 4.64e+02 4.64e+02 2.90e+02 2.90e+02 9.26e-01 8.04e-01 1.0e-03 1000 2.75e+01 3.01e+01 4.40e-02 4.74e-02 8.69e-01 9.51e-01 1.0e-03 2000 2.86e+01 3.14e+01 2.69e-02 4.51e-02 9.03e-01 9.93e-01 1.0e-03 .................. 163000 1.29e+01 1.28e+01 1.86e-02 2.95e-02 9.28e-01 9.14e-01 1.9e-04 164000 1.27e+01 1.42e+01 4.62e-02 4.76e-02 9.07e-01 1.02e+00 1.9e-04 165000 1.33e+01 1.14e+01 2.72e-02 1.08e-02 9.79e-01 8.40e-01 1.8e-04 166000 1.16e+01 1.25e+01 9.23e-02 7.24e-02 8.52e-01 9.21e-01 1.8e-04 167000 1.23e+01 1.14e+01 2.35e-02 6.55e-02 9.04e-01 8.38e-01 1.8e-04 168000 1.20e+01 1.41e+01 4.60e-02 1.97e-02 8.82e-01 1.04e+00 1.8e-04 169000 1.17e+01 1.18e+01 2.20e-02 3.75e-02 8.63e-01 8.65e-01 1.8e-04 170000 1.13e+01 1.00e+01 3.51e-02 9.29e-02 8.55e-01 7.52e-01 1.7e-04 171000 1.22e+01 1.06e+01 3.65e-02 1.23e-02 9.17e-01 7.98e-01 1.7e-04 172000 1.18e+01 1.10e+01 5.21e-02 2.40e-02 8.92e-01 8.29e-01 1.7e-04 173000 1.17e+01 1.13e+01 5.59e-02 2.00e-02 8.83e-01 8.54e-01 1.7e-04 174000 1.24e+01 1.10e+01 7.28e-02 7.33e-02 9.30e-01 8.31e-01 1.7e-04 175000 1.14e+01 1.25e+01 7.05e-02 5.81e-02 8.83e-01 9.64e-01 1.7e-04 176000 1.07e+01 1.23e+01 5.51e-02 2.85e-02 8.24e-01 9.48e-01 1.7e-04 177000 1.13e+01 1.18e+01 7.20e-02 4.19e-02 8.71e-01 9.11e-01 1.7e-04 178000 1.12e+01 1.11e+01 5.49e-02 1.91e-02 8.63e-01 8.60e-01 1.7e-04 179000 1.20e+01 1.04e+01 4.85e-02 3.96e-02 9.28e-01 8.04e-01 1.7e-04 180000 1.19e+01 1.19e+01 6.74e-02 1.78e-02 9.37e-01 9.48e-01 1.6e-04 181000 1.15e+01 1.17e+01 8.61e-02 1.21e-03 9.07e-01 9.26e-01 1.6e-04 182000 1.13e+01 1.04e+01 3.48e-02 1.01e-02 8.94e-01 8.22e-01 1.6e-04 183000 1.05e+01 1.10e+01 3.94e-02 6.62e-02 8.35e-01 8.75e-01 1.6e-04 184000 1.16e+01 1.14e+01 2.73e-02 8.72e-02 9.24e-01 9.01e-01 1.6e-04
Could you please follow the example for the type embedding? Especially the type of the descriptor.
https://github.com/deepmodeling/deepmd-kit/blob/devel/examples/water/se_e2_a_tebd/input.json
For the case of water, one write
"type_embedding":{
"neuron": [2, 4, 8],
"resnet_dt": false,
"seed": 1
},
"descriptor" :{
"type": "se_e2_a",
"sel": [46, 92],
"rcut_smth": 0.50,
"rcut": 6.00,
"neuron": [25, 50, 100],
"resnet_dt": false,
"axis_neuron": 16,
"type_one_side": true,
"seed": 1,
"_comment": " that's all"
},
The type_embedding will automatically be used in the descriptor with type "se_e2_a", which is equivalent to the old one "se_a"
