AutoBADDIE

Autonomous fitting of classical force field parameters

Installation

pip install -e .

Will compile autobaddie into your conda/pip environment. We recommend creating a new environment named autobaddie before running the pip install to avoid conflicts.

Use Tutorial

To use the code, please follow along with the below tutorial. Please open and read through the first sections of the 1. create_template.py, 2. assign_params.py, 3. run_full_training.sh files and update necessary paths and variables. Each of these codes begins with a description of inputs and outputs for each code.

Enter the AutoBADDIE/autobaddie directory and update the create_template.py file.
- The create_template.py code is used to set which atomic environments will be parameterized for a given AutoBADDIE run.
The user must provide a "template_smiles" string representing a set of molecules that encompass all atomic environments for training run that will be referred to by a "job_name". After a successful run of this code, a labelled image for each atom type such as the one below will be created in AutoBADDIE/training_results/job_name/template/atom_types.png. The Lennard-Jones dispersion parameters for each of these atom types will need to be provided in the next code before training.
Update and run the autobaddie/assign_params.py file.
- The user must input a "custom_parameter_dict" which houses the Lennard-Jones parameter for each atomic environment enumerated by the "template_smiles" in the create_template.py file above. After a successful run of this code, a csv file named "template.params" will be saved in the AutoBADDIE/training_results/job_name/template directory. AutoBADDIE will use these Lennard-Jones parameters to optimize all other force field parameters.
Use the run_full_training.sh to take the example training data and learn an AutoBADDIE potential!
- Update the following 3 variables: WORKDIR_BASE, JOB_NAME, DATE
- The job_name must match the previous two examples. The date will be used as a general label to separate this particular optimization run from others for the same chemical system described by the "template_smiles". The training_data_path must contain all the atomic positions, forces, partial charges, and pose energies in a format as displayed in the examples provided (such as AutoBADDIE/training_data/class1).
Once these changes are made, execute the run_full_training.sh code.

If this code is completed successfully, a lammps output for each chemistry escribed in the to_param_list will be saved in the AutoBADDIE/training_results/job_name/date_{EXTRA_TRAINING_DETAILS} directory. Additional information about the training and test performance can be found within this document as well.

License

learningmatter-mit/AutoBADDIE

ErrorLooks like something went wrong!

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Languages

Error
Looks like something went wrong!

Packages