"""
This module provides the StructureContainer class.
"""
import tarfile
import tempfile
from typing import BinaryIO, List, TextIO, Tuple, Union
import numpy as np
import ase.db
from ase import Atoms
from icet import ClusterSpace
from icet.input_output.logging_tools import logger
from pandas import DataFrame
logger = logger.getChild('structure_container')
[docs]
class StructureContainer:
"""This class serves as a container for structure objects as well as their
properties and cluster vectors.
Parameters
----------
cluster_space
Cluster space used for evaluating the cluster vectors.
Example
-------
The following snippet illustrates the initialization and usage of
a :class:`StructureContainer` object. A structure container
provides convenient means for compiling the data needed to train a
cluster expansion, i.e., a sensing matrix and target property values::
>>> from ase.build import bulk
>>> from icet import ClusterSpace, StructureContainer
>>> from icet.tools import enumerate_structures
>>> from random import random
>>> # create cluster space
>>> prim = bulk('Au')
>>> cs = ClusterSpace(prim, cutoffs=[7.0, 5.0],
... chemical_symbols=[['Au', 'Pd']])
>>> # build structure container
>>> sc = StructureContainer(cs)
>>> for structure in enumerate_structures(prim, range(5), ['Au', 'Pd']):
>>> sc.add_structure(structure,
... properties={'my_random_energy': random()})
>>> print(sc)
>>> # fetch sensing matrix and target energies
>>> A, y = sc.get_fit_data(key='my_random_energy')
"""
def __init__(self, cluster_space: ClusterSpace):
if not isinstance(cluster_space, ClusterSpace):
raise TypeError('cluster_space must be a ClusterSpace object.')
self._cluster_space = cluster_space
self._structure_list = []
def __len__(self) -> int:
return len(self._structure_list)
def __getitem__(self, ind: int):
return self._structure_list[ind]
[docs]
def get_structure_indices(self, user_tag: str = None) -> List[int]:
"""
Returns indices of structures with the given user tag. This
method provides a simple means for filtering structures. The
:attr:`user_tag` is assigned when adding structures via the
:func:`add_structure` method.
Parameters
----------
user_tag
The indices of structures with this user tag are returned.
Returns
-------
List of structure indices.
"""
return [i for i, s in enumerate(self) if user_tag is None or s.user_tag == user_tag]
def _get_string_representation(self, print_threshold: int = None) -> str:
"""
String representation of the structure container that provides an
overview of the structures in the container.
Parameters
----------
print_threshold
If the number of structures exceeds this number print dots.
Returns
-------
String representation of the structure container.
"""
if len(self) == 0:
return 'Empty StructureContainer'
# Number of structures to print before cutting and printing dots
if print_threshold is None or print_threshold >= len(self):
print_threshold = len(self) + 2
# format specifiers for fields in table
def get_format(val):
if isinstance(val, float):
return '{:9.4f}'
else:
return '{}'
# table headers
default_headers = ['index', 'user_tag', 'n_atoms', 'chemical formula']
property_headers = sorted(set(key for fs in self for key in fs.properties))
headers = default_headers + property_headers
# collect the table data
str_table = []
for i, fs in enumerate(self):
default_data = [i, fs.user_tag, len(fs), fs.structure.get_chemical_formula()]
property_data = [fs.properties.get(key, '') for key in property_headers]
str_row = [get_format(d).format(d) for d in default_data+property_data]
str_table.append(str_row)
str_table = np.array(str_table)
# find maximum widths for each column
widths = []
for i in range(str_table.shape[1]):
data_width = max(len(val) for val in str_table[:, i])
header_width = len(headers[i])
widths.append(max([data_width, header_width]))
total_width = sum(widths) + 3 * len(headers)
row_format = ' | '.join('{:'+str(width)+'}' for width in widths)
# Make string representation of table
s = []
s += ['{s:=^{n}}'.format(s=' Structure Container ', n=total_width)]
s += ['Total number of structures: {}'.format(len(self))]
s += [''.center(total_width, '-')]
s += [row_format.format(*headers)]
s += [''.center(total_width, '-')]
for i, fs_data in enumerate(str_table, start=1):
s += [row_format.format(*fs_data)]
if i+1 >= print_threshold:
s += [' ...']
s += [row_format.format(*str_table[-1])]
break
s += [''.center(total_width, '=')]
s = '\n'.join(s)
return s
def __str__(self) -> str:
""" String representation. """
return self._get_string_representation(print_threshold=50)
def _repr_html_(self) -> str:
""" HTML representation. Used, e.g., in jupyter notebooks. """
s = ['<h4>Structure Container</h4>']
s += [f'<p>Total number of structures: {len(self)}</p>']
s += self.to_dataframe()._repr_html_()
return ''.join(s)
[docs]
def to_dataframe(self) -> DataFrame:
"""Summary of :class:`StructureContainer` object in :class:`DataFrame
<pandas.DataFrame>` format.
"""
data = []
for s in self:
record = dict(
user_tag=s.user_tag,
natoms=len(s),
formula=s.structure.get_chemical_formula('metal'),
)
record.update(s.properties)
data.append(record)
return DataFrame.from_dict(data)
[docs]
def add_structure(self, structure: Atoms, user_tag: str = None,
properties: dict = None, allow_duplicate: bool = True,
sanity_check: bool = True):
"""
Adds a structure to the structure container.
Parameters
----------
structure
Atomic structure to be added.
user_tag
User tag for labeling structure.
properties
Scalar properties. If properties are not specified the structure
object will be checked for an attached ASE calculator object
with a calculated potential energy.
allow_duplicate
Whether or not to add the structure if there already exists a
structure with identical cluster vector.
sanity_check
Whether or not to carry out a sanity check before adding the
structure. This includes checking occupations and volume.
"""
# structure must have a proper format and label
if not isinstance(structure, Atoms):
raise TypeError(f'structure must be an ASE Atoms object not {type(structure)}')
if user_tag is not None:
if not isinstance(user_tag, str):
raise TypeError(f'user_tag must be a string not {type(user_tag)}.')
if sanity_check:
self._cluster_space.assert_structure_compatibility(structure)
# check for properties in attached calculator
if properties is None:
properties = {}
if structure.calc is not None:
if not structure.calc.calculation_required(structure, ['energy']):
energy = structure.get_potential_energy()
properties['energy'] = energy / len(structure)
# check if there exist structures with identical cluster vectors
structure_copy = structure.copy()
cv = self._cluster_space.get_cluster_vector(structure_copy)
if not allow_duplicate:
for i, fs in enumerate(self):
if np.allclose(cv, fs.cluster_vector):
msg = '{} and {} have identical cluster vectors'.format(
user_tag if user_tag is not None else 'Input structure',
fs.user_tag if fs.user_tag != 'None' else 'structure')
msg += ' at index {}'.format(i)
raise ValueError(msg)
# add structure
structure = FitStructure(structure_copy, user_tag, cv, properties)
self._structure_list.append(structure)
[docs]
def get_condition_number(self, structure_indices: List[int] = None) -> float:
"""Returns the condition number for the sensing matrix.
A very large condition number can be a sign of
multicollinearity. More information can be found
[here](https://en.wikipedia.org/wiki/Condition_number).
Parameters
----------
structure_indices
List of structure indices to include. By default (``None``) the
method will return all fit data available.
Returns
-------
Condition number of the sensing matrix.
"""
return np.linalg.cond(self.get_fit_data(structure_indices, key=None)[0])
[docs]
def get_fit_data(self, structure_indices: List[int] = None,
key: str = 'energy') -> Tuple[np.ndarray, np.ndarray]:
"""
Returns fit data for all structures. The cluster vectors and
target properties for all structures are stacked into numpy arrays.
Parameters
----------
structure_indices
List of structure indices. By default (``None``) the
method will return all fit data available.
key
Name of property to use. If ``None`` do not include property values.
This can be useful if only the fit matrix is needed.
Returns
-------
Cluster vectors and target properties for desired structures.
"""
if structure_indices is None:
cv_list = [s.cluster_vector for s in self._structure_list]
if key is None:
prop_list = None
else:
prop_list = [s.properties[key] for s in self._structure_list]
else:
cv_list, prop_list = [], []
for i in structure_indices:
cv_list.append(self._structure_list[i].cluster_vector)
if key is None:
prop_list = None
else:
prop_list.append(self._structure_list[i].properties[key])
if cv_list is None:
raise Exception(f'No available fit data for {structure_indices}.')
cv_list = np.array(cv_list)
if key is not None:
prop_list = np.array(prop_list)
return cv_list, prop_list
@property
def cluster_space(self) -> ClusterSpace:
""" Cluster space used to calculate the cluster vectors. """
return self._cluster_space
@property
def available_properties(self) -> List[str]:
""" List of the available properties. """
return sorted(set([p for fs in self for p in fs.properties.keys()]))
[docs]
def write(self, outfile: Union[str, BinaryIO, TextIO]) -> None:
"""
Writes structure container to a file.
Parameters
----------
outfile
Output file name or file object.
"""
# Write cluster space to tempfile
temp_cs_file = tempfile.NamedTemporaryFile(delete=False)
self.cluster_space.write(temp_cs_file.name)
# Write fit structures as an ASE db in tempfile
temp_db_file = tempfile.NamedTemporaryFile(delete=False)
temp_db_file.close()
if self._structure_list:
db = ase.db.connect(temp_db_file.name, type='db', append=False)
for fit_structure in self._structure_list:
data_dict = {'user_tag': fit_structure.user_tag,
'properties': fit_structure.properties,
'cluster_vector': fit_structure.cluster_vector}
db.write(fit_structure.structure, data=data_dict)
with tarfile.open(outfile, mode='w') as handle:
handle.add(temp_db_file.name, arcname='database')
handle.add(temp_cs_file.name, arcname='cluster_space')
[docs]
@staticmethod
def read(infile: Union[str, BinaryIO, TextIO]):
"""
Reads :class:`StructureContainer` object from file.
Parameters
----------
infile
File from which to read.
"""
if isinstance(infile, str):
filename = infile
else:
filename = infile.name
if not tarfile.is_tarfile(filename):
raise TypeError('{} is not a tar file'.format(filename))
temp_db_file = tempfile.NamedTemporaryFile(delete=False)
with tarfile.open(mode='r', name=filename) as tar_file:
cs_file = tar_file.extractfile('cluster_space')
temp_db_file.write(tar_file.extractfile('database').read())
temp_db_file.seek(0)
cluster_space = ClusterSpace.read(cs_file)
database = ase.db.connect(temp_db_file.name, type='db')
structure_container = StructureContainer(cluster_space)
fit_structures = []
for row in database.select():
data = row.data
fit_structure = FitStructure(row.toatoms(),
user_tag=data['user_tag'],
cluster_vector=data['cluster_vector'],
properties=data['properties'])
fit_structures.append(fit_structure)
structure_container._structure_list = fit_structures
return structure_container
class FitStructure:
"""
This class holds a supercell along with its properties and cluster
vector.
Attributes
----------
structure
Supercell structure.
user_tag
Custom user tag.
cluster_vector
Cluster vector.
properties
Dictionary comprising name and value of properties.
"""
def __init__(self, structure: Atoms, user_tag: str,
cluster_vector: np.ndarray, properties: dict = {}):
self._structure = structure
self._user_tag = user_tag
self._cluster_vector = cluster_vector
self.properties = properties
@property
def cluster_vector(self) -> np.ndarray:
""" Cluster vector. """
return self._cluster_vector
@property
def structure(self) -> Atoms:
""" Atomic structure. """
return self._structure
@property
def user_tag(self) -> str:
""" Structure label. """
return str(self._user_tag)
def __getattr__(self, key):
""" Accesses properties if possible and returns value. """
if key not in self.properties.keys():
return super().__getattribute__(key)
return self.properties[key]
def __len__(self) -> int:
""" Number of sites in the structure. """
return len(self._structure)
def __str__(self) -> str:
width = 50
s = []
s += ['{s:=^{n}}'.format(s=' Fit Structure ', n=width)]
s += [' {:22} : {}'.format('user tag', self.user_tag)]
for k, v in self.properties.items():
s += [f' {k:22} : {v}']
t = 'cell metric'
for k, row in enumerate(self.structure.cell[:]):
s += [f' {t:22} : {row}']
t = ''
t = 'sites'
for site in self.structure:
s += [f' {t:22} : {site.index} {site.symbol:2} {site.position}']
t = ''
s += [''.center(width, '=')]
return '\n'.join(s)
def _repr_html_(self) -> str:
s = ['<h4>FitStructure</h4>']
s += ['<table border="1" class="dataframe">']
s += ['<thead><tr><th style="text-align: left;">Property</th><th>Value</th></tr></thead>']
s += ['<tbody>']
s += [f'<tr><td style="text-align: left;">user tag</td><td>{self.user_tag}</td></tr>']
for key, value in sorted(self.properties.items()):
s += [f'<tr><td style="text-align: left;">{key}</td><td>{value}</td></tr>']
s += ['</tbody></table>']
s += ['<table border="1" class="dataframe">']
s += ['<thead><tr><th style="text-align: left;">Cell</th></tr></thead>']
s += ['<tbody>']
for row in self.structure.cell[:]:
s += ['<tr>']
for c in row:
s += [f'<td>{c}</td>']
s += ['</tr>']
s += ['</tbody></table>']
df = DataFrame(np.array([self.structure.symbols,
self.structure.positions[:, 0],
self.structure.positions[:, 1],
self.structure.positions[:, 2]]).T,
columns=['Species', 'Position x', 'Position y', 'Position z'])
s += df._repr_html_()
return ''.join(s)