ALFF: Active Learning

Run the main ALFF active learning process. The active learning process in ALFF is designed to perform the following tasks automatically and iteratively without needs any user intervention:

1. Train ML models

   • Collect data files
   • Split dataset
   • Prepare training args
   • Training ML models at remote machines
   • Monitor the training job status
   • Get back the trained ML models whenever they are finished

2. Run MD simulations for sampling exploration

   • Configure sampling exploration spaces
   • Prepare MD args
   • Submit and run MD simulations to explore atomic configurations at remote machines
   • Monitor the MD job status and get back the results whenever they are finished
   • Select condidate atomic configurations for DFT calculations

3. Run DFT calculations for labeling the selected atomic configurations

   • Prepare DFT args and DFT tasks
   • Submit and run DFT jobs to remote machines
   • Monitor the job status, and get back the DFT calculation results
   • Collect the data from the DFT calculations
   • Convert labeled data to the readable formats for training ML models on the next iteration.

alff_al PARAM.yml MACHINE.yml

• PARAM.yml: The parameters of the generator.
• MACHINE.yml: The settings of the machines running the generator's subprocesses.

An example run:

• 0^th iteration:

• 1^st iteration:

• 2^nd iteration:

• Nth iteration: