preprocessing

Data formats used in this package: I) The .npz format is a file format used by NumPy. It can store multiple arrays within a single .npz file, making it convenient for bundling related data together. Each array in the .npz file is associated with a unique key, which can be used to access the specific array when loading the data.

II) The Extended XYZ (.xyz) format is a file format commonly used for representing molecular structures and atomic coordinates. It provides a simple and flexible way to store atomic positions, element types, and additional properties associated with each atom. Each frame can represent a different time step, molecular conformation, or related structure.

NPZ and XYZ can be interconverted b/w each other.

Information needed:

1. 'atomic_numbers'
2. 'cell'
3. 'forces'
4. 'energy'
5. 'positions'
6. 'pbc' #preriodic boundary condition in all 3 directions

example of accepted data: present at example/lips/data =============Lips data info============ Keys for data: key: pbc with shape (3,) key: pos with shape (1000, 83, 3) key: energy with shape (1000, 1) key: forces with shape (1000, 83, 3) key: cell with shape (3, 3) key: atomic_numbers with shape (83,)

NOTE:

1. Nequip, Allegro, Equiformer and TorchmdNET can work with just atomic numbers, forces, energy and positions.
2. By default the models use the xyz file name as botnet.xyz and npz as nequip_npz.npz

All the models: Nequip, Allegro, Equiformer, torchmdNET, Mace and Botnet support both npz and xyz format for training

converting XYZ to NPZ: run file xyz_to_npz

CONVERTING NPZ to XYZ: example: python preprocessing/npz_to_xyz.py --input-filepath example/lips/data/val/nequip_npz.npz will create an xyz file at example/lips/data/val/botnet.xyz

IMP NOTE: In case facing some error in data conversion:

1. get data in npz format and cross check with example/lips/data/train/nequip_npz.npz
2. Then create test, train and val split like we have here example/lips/data
3. Then convert npz to xyz