5.4. [train] section

abics_train creates and trains a regression model from configurations to energies. Indeed, abics_train uses an external program to train the model. In the current version, Aenet, Nequip, and MLIP-3 are supported as an external program. For software-specific notes (such as input file names), see Machine learning trainer/calculator-specific notes.

The input information for abics_train is described in the [trainer] section. The description of each parameter is as follows. An example is shown as follows:

[trainer] # Configure the model trainer.
type = 'aenet'
base_input_dir = '. /aenet_train_input'
exe_command = ['~/git/aenet/bin/generate.x-2.0.4-ifort_serial',
             'srun ~/git/aenet/bin/train.x-2.0.4-ifort_intelmpi']
ignore_species = ["O"]

5.4.1. Input Format

Keywords and their values are specified by a keyword and its value in the form keyword = value. Comments can also be entered by adding # (Subsequent characters are ignored).

5.4.2. Key words

  • type

    Format : str

    Description : The trainer to generate the neural network potential (currently ‘aenet’, ‘nequip’, and ‘mlip_3’ are available).

  • base_input_dir

    Format : str

    Description : Path of the directory containing the input files that the learner refers to.

  • exe_command

    Format : dict

    Description : List of commands to execute; if you use aenet, you need to specify the path to generate.x and train.x.

    • type = 'aenet'

      • generate and train keys are required.

      • generate

        • Specify the path to generate.x of aenet.

      • train

        • Specify the path to train.x of aenet.

        • The MPI parallel version is available. In that case, set the command to execute MPI (e.g., srun, mpirun) .

      • Array format is supported for compatibility with abICS 2.0 and earlier. The first element is generate, and the second element is train.

    • type = 'nequip'

      • train

        • Specify the path to nequip-train.

    • type = 'mlip_3'

      • train

        • Specify the path to mlp.

  • ignore_species

    Format : list

    Description : Same as ignore_species in [sampling.solver] section. Specify atomic species to “ignore” in neural network models such as aenet. For those that always have an occupancy of 1, it is computationally more efficient to ignore their presence when training and evaluating neural network models.