featurebox.featurizers.envir package¶

This part is for local environment features, mainly contains bond and state features.

Subpackages¶

featurebox.featurizers.envir.descriptors package

Submodules¶

featurebox.featurizers.envir.desc_env module¶

Use descriptors form pyXtal_FF, in featurebox.test_featurizers.descriptors all with calculate method.

featurebox.featurizers.envir.desc_env.get_5_result(d: Dict, **kwargs) → Tuple[ndarray, ndarray, ndarray, ndarray, ndarray]¶

Change each center atoms has fill_size neighbors. More neighbors would be abandoned. Insufficient neighbors would be duplicated.

Parameters:

d – dict, dict of descriptor. at lest contain “x” and “dxdr”
fill_size – int, unstable.

Returns:

(center_indices,center_prop, neighbor_indices, images, distances)

center_indices: np.ndarray 1d(N,).

neighbor_indices: np.ndarray 2d(N,fill_size).

images: np.ndarray 2d(N,fill_size,l).

distance: np.ndarray 2d(N,fill_size), None. center_prop: np.ndarray 1d(N,l_c).

featurebox.featurizers.envir.desc_env.get_strategy2_in_spheres(structure, nn_strategy, cutoff, numerical_tol=None, cutoff_name='cutoff', pbc=True)¶

featurebox.featurizers.envir.desc_env.mark_classes(classes: List)¶

featurebox.featurizers.envir.environment module¶

class featurebox.featurizers.envir.environment.GEONNGet(nn_strategy: str = 'UserVoronoiNN', refine: str = 'geo_refine_nn', refined_strategy_param: Optional[Dict] = None, numerical_tol=1e-08, pbc: Union[List[int], bool] = False, cutoff=5.0, check_align=True, cutoff_name='cutoff', n_jobs=1, on_errors='raise', return_type='any')¶

Bases: _BaseEnvGet

Get properties from Pymatgen.Structure. Where the nn_strategy is from Pymatgen. And each structure is convert to data as following :

center_indices:np.ndarray of shape(n,): center indexes.
neighbor_indices:np.ndarray of shape(n,fill_size): neighbor_indexes for each center_index. fill_size is the parameter of refine function.
images:np.ndarray of shape(n,lb>=3): offset vector in 3 orientations or more bond properties.
distances:np.ndarray of shape(n,fill_size): distance of neighbor_indexes for each center_index.
center_properties:np.ndarray of shape(n,l_c): center properties.

Parameters:

nn_strategy (str) – [“find_points_in_spheres”, “find_xyz_in_spheres”, “BrunnerNN_reciprocal”, “BrunnerNN_real”, “BrunnerNN_relative”, “EconNN”, “CrystalNN”, “MinimumDistanceNNAll”, “find_points_in_spheres”,”UserVoronoiNN”, “ACSF”,”BehlerParrinello”,”EAD”,”EAMD”,”SOAP”,”SO3”,”SO4_Bispectrum”,”wACSF”,]
refine (str) – sort method for neighbors of each atom. See Also: universe_refine_nn()
refined_strategy_param (dict) – parameters for refine
numerical_tol (float) – numerical_tol
pbc (list) – periodicity in 3 direction 3-length list,each one is 1 or 0. such as [0,0,0],The 1 mean in this direction is with periodicity.
cutoff – cutoff radius

convert(structure: Union[Structure, Molecule]) → Tuple[ndarray, ndarray, ndarray, ndarray, ndarray]¶

Parameters:: structure – Structure,pymatgen Structure
Returns:: center_indices,center_prop,neighbor_indices,images,distances

get_radius(structure: Union[Structure, Molecule], cutoff) → Tuple[ndarray, ndarray, ndarray, ndarray, ndarray]¶: For quick get bond distance

get_strategy1(structure: Union[Structure, Molecule], cutoff) → Tuple[ndarray, ndarray, ndarray, ndarray, ndarray]¶: For get bond distance with different strategy, for different nn_staagy could be rewrite.

get_strategy2(structure: Union[Structure, Molecule], cutoff) → Tuple[ndarray, ndarray, ndarray, ndarray, ndarray]¶: For get bond distance with different strategy, for different nn_staagy could re-write.

get_xyz(structure: Union[Structure, Molecule], cutoff) → Tuple[ndarray, ndarray, ndarray, ndarray, ndarray]¶: For quick get bond distance

featurebox.featurizers.envir.environment.geo_refine_nn(center_indices, neighbor_indices, vectors, distances, center_prop=None, ele_numbers=None, fill_size=10, dis_sort=True, **kwargs)¶

Change each center atoms has fill_size neighbors. More neighbors would be abandoned. Insufficient neighbors would be duplicated.

Parameters:

center_indices – np.ndarray 1d
neighbor_indices – np.ndarray 1d
distances – np.ndarray 1d or np.ndarray 2d
vectors – np.ndarray 2d
center_prop – np.ndarray 2d
ele_numbers – np.ndarray 1d
fill_size – float, not use,
dis_sort – bool sort neighbors with distance.

Returns:

(center_indices,center_indices, neighbor_indices, images, distances)

center_indices: np.ndarray 1d(N,).

neighbor_indices: np.ndarray 2d(N,fill_size).

images: np.ndarray 2d(N,fill_size,l).

distance: np.ndarray 2d(N,fill_size,1). center_prop: np.ndarray 1d(N,l_c).

where l, and l_c >= 1

featurebox.featurizers.envir.environment.get_marked_class(nn_strategy, env_dict: Optional[Dict] = None, instantiation: bool = True)¶

Parameters:

nn_strategy (Any) – “find_points_in_spheres”, “find_xyz_in_spheres” , or nn_strategy
env_dict (dict) – pre-definition, {“classname”: class}.
instantiation (bool) – return class of object.

Returns:

object or class in NNDict.

Return type:

obj

featurebox.featurizers.envir.local_env module¶

Use NearNeighbors from pymatgen. all with get_all_nn_info method.

Most in pymatgen.analysis.local_env or in pymatgen.optimization.neighbors.find_points_in_spheres() The costumed as following:

class featurebox.featurizers.envir.local_env.AllAtomPairs¶

Bases: NearNeighbors

Get all combinations of atoms as bonds in a molecule

get_nn_info(molecule: Molecule, n: int) → List[Dict]¶

Get near neighbor information :param molecule: pymatgen Molecule :type molecule: Molecule :param n: number of molecule :type n: int

Returns: List of neighbor dictionary

class featurebox.featurizers.envir.local_env.MinimumDistanceNNAll(cutoff: float = 4.0)¶

Bases: NearNeighbors

Determine bonded sites by fixed cutoff.

Parameters:: cutoff (float) – cutoff radius in Angstrom to look for trial near-neighbor sites (default: 4.0).

get_nn_info(structure: Structure, n: int) → List[Dict]¶

Get all near-neighbor sites as well as the associated image locations and weights of the site with index n using the closest neighbor distance-based method.

Parameters:

structure (Structure) – input structure.
n (int) – index of site for which to determine near neighbors.

Returns:

tuples, each one of which represents a neighbor site, its image location,: and its weight.

Return type:

(list of tuples (Site, array, float))

class featurebox.featurizers.envir.local_env.UserVoronoiNN(tol=0, targets=None, cutoff=13.0, allow_pathological=False, weight='solid_angle', extra_nn_info=True, compute_adj_neighbors=True)¶

Bases: VoronoiNN

Not for all structure.

Parameters:

tol (float) – tolerance parameter for near-neighbor finding. Faces that are smaller than tol fraction of the largest face are not included in the tessellation. (default: 0).
targets (Element or list of Elements) – target element(s).
cutoff (float) – cutoff radius in Angstrom to look for near-neighbor atoms. Defaults to 13.0.
allow_pathological (bool) – whether to allow infinite vertices in determination of Voronoi coordination.
weight (string) – available in get_voronoi_polyhedra)
extra_nn_info (bool) –
compute_adj_neighbors (bool) – adjacent. Turn off for faster performance

featurebox.featurizers.envir.local_env.get_strategy1_in_spheres(structure: Union[Structure, Molecule], nn_strategy: NearNeighbors, cutoff: float = 5.0, numerical_tol: float = 1e-08, pbc=True) → Tuple[ndarray, ndarray, ndarray, ndarray, ndarray]¶

featurebox.featurizers.envir.local_env.mark_classes(classes: List)¶