set

The dptools set command is one of the most used commands in DPTools! It is used to save various system configurations:

  • deepmd-kit models (.pb files)

  • HPC (slurm) parameters (.sh files)

  • simulation parameters (.yaml files)

  • deepmd training parameters (.json files)

General usage,

$ dptools set [-h] [-m MODEL_LABEL] thing [thing ...]

positional arguments:
  thing                 Path to DP model, params.yaml, or {script}.sh to set as default. Need .pb, .yaml, or .sh extension to set model, params, or sbatch, respectively.

optional arguments:
  -h, --help            show this help message and exit
  -m MODEL_LABEL, --model-label MODEL_LABEL
                        Save model/sbatch with specific label to access during 'dptools run -m {label} {calc} {structure}' (default: None)

Note

To see information on all set models and sbatch parameters, use the dptools info command.

Quick reference examples

$ dptools set /home/user/projects/dp_models/h2o.pb
$ dptools set 00/graph.pb 01/graph.pb 02/graph.pb 03/graph.pb # ensemble
$ dptools set -m water /home/user/projects/dp_models/h2o.pb
$ dptools set -m hpc_para parallel_settings.sh
$ dptools set -m hpc_serial serial_settings.sh
$ dptools set params.yaml
$ dptools set /path/to/in.json

Set default DP model

Setting a DP model (.pb tensorflow graph file) allows you to run a simulation. Just supply the path to the model file,

$ dptools set /path/to/graph.pb

This model will now be used as the default for dptools run.

Set labelled DP model

If you’re working with multiple DP models (e.g. for two different projects with different systems or for comparing two different model architectures for the same system), you can set additional models with a specific label.

$ dptools set -m label /path/to/graph.pb

The label can now be used to specify the model when you run simulations, like for example,

$ dptools run -m label opt start.traj

Set ensemble of DPs

If you want to set an ensemble of model to use for sampling new configurations for training, you can set the ensemble by specifying the paths to each DP model,

$ dptools set 00/graph.pb 01/graph.pb 02/graph.pb 03/graph.pb

You can also set a label for your ensemble in the same way you do for a single model,

$ dptools set -m iteration2_ensemble 00/graph.pb 01/graph.pb 02/graph.pb 03/graph.pb

Note

You must specify all model files in a single command to set an ensemble. If you try to set them individually like,

$ dptools set 00/graph.pb # don't do this!
$ dptools set 01/graph.pb # don't do this!

The 01/graph.pb is not appended to a list of models! You are simply replacing 00/graph.pb with 01/graph.pb.

Modify default simulation parameters

To change the default simulation parameters for any simulation type, retrieve the corresponding params.yaml file with the get command. To modify the nvt-md parameters for example,

$ dptools get nvt-md

Edit the params.yaml file in whatever text editor you prefer, and then set the file,

$ dptools set params.yaml

These new parameters will now be used everytime you run the simulation,

$ dptools run nvt-md start.traj

Create custom simulation parameter sets

This is useful if you want to store multiple parameter settings for the same type of simulation so you don’t have to constantly retrieve and modify params.yaml files if you’ll be using these settings frequently.

For example, let’s say you want to store several different temperature settings for an NVT-MD simulation. First get the params.yaml file for the nvt-md simulation,

$ dptools get nvt-md

Opening the params.yaml file will look something like,

type: nvt-md        # Type of calculation (spe, opt, cellopt, nvt-md, npt-md, eos)
steps: 100000       # Total number of timesteps to run simulation
timestep: 0.5       # [fs]
Ti: 298.0           # Initial temperature [K] at start of simulation
Tf: 298.0           # Final temperature [K] of simulation (ramped from Ti to Tf)
equil_steps: 10000  # Number of timesteps to run initial equilibration at Ti
write_freq: 100     # Write MD image every {write_freq} steps
disp_freq: 100      # Print lammps output every {disp_freq} steps (thermo disp_freq)
pre_opt: false      # Optimize structure (and cell for npt-md) before starting MD run

To create a custom parameter set for type: nvt-md, modify the type name following the format type: nvt-md.label. You must retain the original simulation type name before the appended label!

Let’s make a new set of parameters for 600 K by modifying the params.yaml file like so,

type: nvt-md.600K   # Type of calculation (spe, opt, cellopt, nvt-md, npt-md, eos)
steps: 100000       # Total number of timesteps to run simulation
timestep: 0.5       # [fs]
Ti: 600.0           # Initial temperature [K] at start of simulation
Tf: 600.0           # Final temperature [K] of simulation (ramped from Ti to Tf)
equil_steps: 10000  # Number of timesteps to run initial equilibration at Ti
write_freq: 100     # Write MD image every {write_freq} steps
disp_freq: 100      # Print lammps output every {disp_freq} steps (thermo disp_freq)
pre_opt: false      # Optimize structure (and cell for npt-md) before starting MD run

Now we can save the file and set it as usual,

$ dptools set params.yaml

And the simulation is now available for use directly with the run command!

$ dptools run nvt-md.600K start.traj

To expand on this example and demonstrate why this might be useful, let’s repeat this for 400 K and 800 K. Follow the same steps above, and now when we run dptools get list, we will see our new custom simulations,

$ dptools get list

# Available calculation types:
spe
opt
cellopt
nvt-md
npt-md
eos
vib
nvt-md.600K
nvt-md.400K
nvt-md.800K

Now say we have several systems, such as 3 different metal-organic frameworks (which probably wouldn’t be stable at 800 K, but we’ll ignore that), that we want to run separate NVT-MD simulations at these 3 temperatures. Our folder structure might look something like:

.
|-- 00_mof0
|   |-- 400
|   |   `-- mof0_start.traj
|   |-- 600
|   |   `-- mof0_start.traj
|   `-- 800
|       `-- mof0_start.traj
|-- 01_mof1
|   |-- 400
|   |   `-- mof1_start.traj
|   |-- 600
|   |   `-- mof1_start.traj
|   `-- 800
|       `-- mof1_start.traj
`-- 02_mof2
    |-- 400
    |   `-- mof2_start.traj
    |-- 600
    |   `-- mof2_start.traj
    `-- 800
        `-- mof2_start.traj

Note that all mof0_start.traj are equivalent structures (same for mof1 and mof2), we just need a separate file in each directory to pass as an argument to dptools run (#TODO: modify things to allow for multiple simulations stemming from a single input structure file).

Assuming we have already set a model and Slurm settings. We are now ready to submit these simulations to run! Simply use these 3 commands specifying our custom temperature simulations:

$ dptools run -s nvt-md.400K 0*/400/*start.traj
$ dptools run -s nvt-md.600K 0*/600/*start.traj
$ dptools run -s nvt-md.800K 0*/800/*start.traj

Set Slurm parameters

Set new training parameters

To change the default deepmd-kit training parameters, first retreive the in.json file using the get command,

$ dptools get in.json

Edit the file in whatever text editor you prefer, and then set the file,

$ dptools set in.json