How to apply bamboo to your own dataset?

[loilisxka]

Hi, I am studying the use of machine learning force fields for battery cathode material simulation. I would like to ask if you can provide a bamboo python training script. I want to use my own dataset for training and use ASE for molecular dynamics simulation, so I would like to ask if you can provide a pure python bamboo training script. All codes will be used for academic research and not for any commercial purpose.

[samueldyoung29ctr]

Are you asking how to invoke the BAMBOO training script from within Python rather than using python3 -m train.train --config configs/train_config/config.json ... on the command line? I would think you just have to make the train/train.py file available on your $PYTHONPATH, then import the train.BambooTrainer class, create the default arguments, override with any of your own, and run the training process. Something like this:

import sys
sys.path.append("<path to BAMBOO repository>")

from bamboo.train import BambooTrainer, get_parser

# Get default args.
args = get_parser()

# Override default arguments with any of your own.
args.rcut = 6.0
args.num_layers = 2

# Create trainer class/
bamboo_trainer = BambooTrainer(args)

# Run training process.
bamboo_trainer.train(args.num_epoch)