Data Representation

mlcg.data defines the AtomicData class, the data structure used as input to models. It contains one or more atomic structures with associated physical properties. AtomicData instances also store the outputs of models through which they pass.

class mlcg.data.atomic_data.AtomicData(**kwargs)[source]

A data object holding atomic structures. The attribute names are defined in mlcg.data._keys

pos

set of atomic positions in each structures

Type:: [n_atoms * n_structures, 3]

atom_types

if atoms then it’s the atomic number, if it’s a CG bead then it’s a number defined by the CG mapping

Type:: [n_atoms * n_structures]

masses

Masses of each atom

Type:: [n_atoms * n_structures]

pbc

periodic boundary conditions

Type:: [n_structures, 3] (Optional)

cell

unit cell of the atomic structure. Lattice vectors are defined row wise

Type:: [n_structures, 3, 3] (Optional)

tag

metadata about each structure

Type:: [n_structures] (Optional)

energy

reference energy associated with each structures

Type:: [n_structures] (Optional)

forces

reference forces associated with each structures

Type:: [n_atoms * n_structures, 3] (Optional)

velocities

velocities associated with each structure

Type:: [n_atoms * n_structures, 3] (optional)

neighborlist

contains information about the connectivity formatted according to mlcg.neighbor_list.neighbor_list.make_neighbor_list.

Type:: Dict[str, Dict[str, Any]] (Optional)

batch

maps the atoms to their structure index (from 0 to n_structures-1)

Type:: [n_atoms]