Hands-on tutorials: DeePMD-kit

Example download here.
Manual download here.
For each specific example, we provide detailed descriptions and instructions within the corresponding sub-folder:

cd example
ls
example1_intro_tf         example3_train_input  example5_lmp_eam
example2_fit_sin_with_nn  example4_train_water  example6_workflow

You may go through the instrcutions in each example or follow the instructions below.

Example 1-2: basics of TensorFlow and Deep Learning

  1. Go to Example 1 and run the python code:
cd example1_intro_tf
python Introduction+to+TensorFlow.py
  1. Go to Example 2 and run the python code:
cd ../example2_fit_sin_with_nn/
python Fitting+sine+curve.py 

You might need to close the figure to continue with the code.

Example 3-5: Training and MD

  1. Go to Example 3:
cd ../example3_train_input/

You may freely have a look at input files for various different systems – they are highly similar!

For example,

cd deeppot_se
vi HEA.json
cd ..
  1. Go to Example 4 and run a training code:
cd ../example4_train_water/train
dp_train water.json

Things like this might appear:

#
# find 1 system(s): 
# find system ../data/water/ :      192 atoms            75 batches copied by [1]
#
# run with intra_op_parallelism_threads = 1, inter_op_parallelism_threads = 0 
2018-08-02 12:32:03.411340: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
# computed stats
# initialize model from scratch
# start training, start lr is 1.000000e-03, final lr will be 3.505267e-08
2018-08-02 12:32:04.760812: W tensorflow/core/framework/allocator.cc:101] Allocation of 519782400 exceeds 10% of system memory.
2018-08-02 12:32:12.163116: W tensorflow/core/framework/allocator.cc:101] Allocation of 519782400 exceeds 10% of system memory.
# batch     100 training time 6.02 s, testing time 1.25 s
# batch     100 training time 6.02 s, testing time 1.25 s
# saved checkpoint /home/dft003/deepmd-kit-tutorial/example/example4_train_water/train/model.ckpt
2018-08-02 12:32:19.596083: W tensorflow/core/framework/allocator.cc:101] Allocation of 519782400 exceeds 10% of system memory.
# batch     200 training time 5.92 s, testing time 1.26 s
......

More instructions can be found in the section $3.2 Train a model $ in the file manual.pdf contained in deepmd-kit-tutorialfolder.

  1. Go to Example 5:
cd ../../example5_lmp_eam/
ls
Al_mm.eam.fs  diffusion  equiv  melt  structure_factor  surf_energy

You may have a look at input files for various different MD purposes.

For example, if you want to find the melting point of Al under the EAM potential, you may do the following:

cd melt
lmp_serial < in.melt

For other purposes, go to the corresponding folder, and do lmp_serial < in.xxx, where in.xxx is the corresponding input file.

We cannot show in more details how to run LAMMPS, which is not our purpose here.

We refer to ~/deepmd-kit/soft/lammps-16Mar18/examples/ for more examples.

Example 6: Workflow for new studies

  1. Go to Example 6:
cd ../../example6_workflow/

Do data conversion:

python xyz2npyplus.py

Go to the training folder and do training:

cd train
dp_train input.json