# Copyright 2021 IRT Saint Exupéry, https://www.irt-saintexupery.com
#
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU Lesser General Public
# License version 3 as published by the Free Software Foundation.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# Lesser General Public License for more details.
#
# You should have received a copy of the GNU Lesser General Public License
# along with this program; if not, write to the Free Software Foundation,
# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
# Contributors:
# INITIAL AUTHORS - initial API and implementation and/or initial
# documentation
# :author: Francois Gallard, Charlie Vanaret
# OTHER AUTHORS - MACROSCOPIC CHANGES
"""A function computing some outputs of a discipline from some of its inputs."""
from __future__ import annotations
from typing import TYPE_CHECKING
from numpy import array
from numpy import empty
from numpy import ndarray
from scipy.sparse import block_array
from scipy.sparse import block_diag
from scipy.sparse import csr_array
from scipy.sparse import dok_array
from gemseo.core.execution_status import ExecutionStatus
from gemseo.core.mdo_functions.mdo_function import MDOFunction
from gemseo.utils.compatibility.scipy import get_row
from gemseo.utils.compatibility.scipy import sparse_classes
from gemseo.utils.constants import READ_ONLY_EMPTY_DICT
if TYPE_CHECKING:
from collections.abc import MutableMapping
from collections.abc import Sequence
from gemseo.core.discipline import Discipline
from gemseo.core.grammars.grammar_properties import GrammarProperties
from gemseo.typing import JacobianData
from gemseo.typing import NumberArray
from gemseo.typing import StrKeyMapping
[docs]
class DisciplineAdapter(MDOFunction):
"""An :class:`.MDOFunction` executing a discipline for some inputs and outputs."""
__is_linear: bool
"""Whether the function is linear."""
__input_dimension: int | None
"""The input variable dimension, needed for linear candidates."""
differentiated_input_names_substitute: Sequence[str]
"""The names of the inputs with respect to which to differentiate the functions.
If empty, consider the variables of their input space.
"""
def __init__(
self,
input_names: Sequence[str],
output_names: Sequence[str],
default_input_data: GrammarProperties,
discipline: Discipline,
names_to_sizes: MutableMapping[str, int] = READ_ONLY_EMPTY_DICT,
differentiated_input_names_substitute: Sequence[str] = (),
) -> None:
"""
Args:
input_names: The names of the inputs.
output_names: The names of the outputs.
default_input_data: The default input values
to overload the ones of the discipline
at each evaluation of the outputs with :meth:`._fun`
or their derivatives with :meth:`._jac`.
If empty, do not overload them.
discipline: The discipline to be adapted.
names_to_sizes: The sizes of the input variables.
If empty, determine them from the default inputs and local data
of the discipline :class:`.Discipline`.
differentiated_input_names_substitute: The names of the inputs
with respect to which to differentiate the functions.
If empty, consider the variables of their input space.
""" # noqa: D205, D212, D415
super().__init__(
self._func_to_wrap,
jac=self._jac_to_wrap,
name="_".join(output_names),
input_names=input_names,
output_names=output_names,
)
self.differentiated_input_names_substitute = (
differentiated_input_names_substitute or input_names
)
self.__default_inputs = default_input_data
self.__input_size = 0
self.__differentiated_input_size = 0
self.__jacobian = array(())
self.__discipline = discipline
self.__input_names_to_slices = {}
self.__input_names_to_sizes = names_to_sizes or {}
self.__differentiated_input_names_to_slices = {}
input_names = set(self.input_names)
self.__is_linear = self.__discipline.io.have_linear_relationships(
input_names, output_names
)
self.__input_dimension = self.__compute_input_dimension(default_input_data)
self.__convert_array_to_data = (
discipline.io.input_grammar.data_converter.convert_array_to_data
)
@property
def is_linear(self) -> bool: # noqa: D102
return self.__is_linear
@property
def input_dimension(self) -> int | None: # noqa: D102
return self.__input_dimension
def __compute_input_dimension(
self,
default_input_data: GrammarProperties,
) -> int | None:
"""Compute the input dimension.
Args:
default_input_data: : The default input values
to overload the ones of the discipline
at each evaluation of the outputs with :meth:`._fun`
or their derivatives with :meth:`._jac`.
If ``None``, do not overload them.
Returns:
The input dimension.
"""
get_value_size = (
self.__discipline.io.input_grammar.data_converter.get_value_size
)
if default_input_data and all(
name in default_input_data for name in self.input_names
):
return sum(
get_value_size(input_name, default_input_data[input_name])
for input_name in self.input_names
)
if len(self.__input_names_to_sizes) > 0:
return sum(self.__input_names_to_sizes.values())
default_input_data = self.__discipline.io.input_grammar.defaults
if all(name in default_input_data for name in self.input_names):
return sum(
get_value_size(input_name, default_input_data[input_name])
for input_name in self.input_names
)
# TODO: document what None means. We could use 0 instead.
return None
def _func_to_wrap(self, x_vect: NumberArray) -> complex | NumberArray:
"""Compute an output vector from an input one.
Args:
x_vect: The input vector.
Returns:
The output vector or a scalar if the vector has only one component.
"""
self.__discipline.execution_status.value = ExecutionStatus.Status.DONE
input_data = self.__create_discipline_input_data(x_vect)
n_samples = len(x_vect) if x_vect.ndim == 2 else 1
data = self.__discipline.execute(input_data)
return self._convert_output_data_to_array(
{k: v for k, v in data.items() if k in self.output_names},
n_samples=n_samples,
)
def _convert_output_data_to_array(
self, output_data: StrKeyMapping, n_samples: int = 1
) -> complex | NumberArray:
"""Convert the discipline's output data to array/scalar.
Args:
output_data: The discipline's output data.
n_samples: The number of samples.
Returns:
The vector or scalar of output data.
"""
if n_samples > 1:
output_data = {
k: v.reshape((n_samples, -1)) if isinstance(v, ndarray) else v
for k, v in output_data.items()
}
output_vector = (
self.__discipline.io.output_grammar.data_converter.convert_data_to_array(
self.output_names, output_data
)
).ravel()
if output_vector.size == 1: # The function is scalar.
return output_vector[0]
return output_vector
def _jac_to_wrap(self, x_vect: NumberArray) -> NumberArray:
"""Compute the Jacobian value from an input vector.
Args:
x_vect: The input vector.
Returns:
The Jacobian value.
"""
input_data = self.__create_discipline_input_data(x_vect)
n_samples = len(x_vect) if x_vect.ndim == 2 else 1
jacobian_data = self.__discipline.linearize(input_data)
return self._convert_jacobian_to_array(jacobian_data, n_samples=n_samples)
def _convert_jacobian_to_array(
self, jacobian_data: JacobianData, n_samples: int = 1
) -> NumberArray:
"""Convert the discipline's Jacobians to array.
Args:
jacobian_data: The discipline's Jacobian data.
n_samples: The number of samples.
Returns:
The aggregated Jacobian as a NumPy array.
"""
if n_samples > 1 or len(self.__jacobian) == 0:
self.__initialize_jacobian(jacobian_data, n_samples)
# Case 1 = "1 sample and output_dimension = 1".
if self.__jacobian.ndim == 1 or self.__jacobian.shape[0] == 1:
self.__convert_jacobian_to_array_case_1(jacobian_data)
return self.__jacobian
# Case 2 = "1 sample and output_dimension > 1".
if n_samples == 1:
self.__convert_jacobian_to_array_case_2(jacobian_data)
return self.__jacobian
# Case 3 = "n samples".
self.__convert_jacobian_to_array_case_3(jacobian_data, n_samples)
return self.__jacobian
def __convert_jacobian_to_array_case_1(self, jacobian_data: JacobianData) -> None:
"""Convert the Jacobian data in the case "1 sample and output_dimension = 1".
Args:
jacobian_data: The discipline's Jacobian data.
"""
jacobian_ = self.__jacobian
jacobian_data_ = jacobian_data[self.output_names[0]]
for input_name in self.differentiated_input_names_substitute:
input_slice = self.__differentiated_input_names_to_slices[input_name]
# jacobian_data__ is a (n*pi, n*dj) array
jacobian_data__ = jacobian_data_[input_name]
# TODO: This precaution is meant to disappear when sparse 1-D array will
# be available. This is also mandatory since self.__jacobian is
# initialized as a dense array.
if isinstance(jacobian_data__, sparse_classes):
first_row = get_row(jacobian_data__, 0).todense().flatten()
else:
first_row = jacobian_data__[0, :]
jacobian_[input_slice] = first_row
def __convert_jacobian_to_array_case_2(self, jacobian_data: JacobianData) -> None:
"""Convert the Jacobian data in the case "output_dimension > 1".
Args:
jacobian_data: The discipline's Jacobian data.
"""
jacobian_ = self.__jacobian
output_position = 0
for output_name in self.output_names:
jacobian_data_ = jacobian_data[output_name]
input_position = 0
output_dimension = 0
for input_name in self.differentiated_input_names_substitute:
jacobian_data__ = jacobian_data_[input_name]
# TODO: This is mandatory since self.__jacobian is initialized as a
# dense array. Performance improvement could be obtained if one is
# able to infer the type of jac.
if isinstance(jacobian_data__, sparse_classes):
jacobian_data__ = jacobian_data__.toarray()
output_dimension, input_dimension = jacobian_data__.shape
jacobian_[
output_position : output_position + output_dimension,
input_position : input_position + input_dimension,
] = jacobian_data__
input_position += input_dimension
output_position += output_dimension
def __convert_jacobian_to_array_case_3(
self, jacobian_data: JacobianData, n_samples: int
) -> None:
"""Convert the Jacobian data in the case "n samples or output_dimension > 1".
Args:
jacobian_data: The discipline's Jacobian data.
n_samples: The number of samples.
"""
n_inputs = len(self.differentiated_input_names_substitute)
n_outputs = len(self.output_names)
output_names_to_sizes = {
k: next(iter(v.values())).shape[0] // n_samples
for k, v in jacobian_data.items()
}
diagonal_arrays = []
for i in range(n_samples):
arrays = [[None] * n_inputs] * n_outputs
for output_index, output_name in enumerate(self.output_names):
jacobian_data_ = jacobian_data[output_name]
output_size = output_names_to_sizes[output_name]
arrays_ = arrays[output_index]
for input_index, input_name in enumerate(
self.differentiated_input_names_substitute
):
jacobian_data__ = jacobian_data_[input_name]
input_size = self.__input_names_to_sizes[input_name]
arrays_[input_index] = jacobian_data__[
i * output_size : (i + 1) * output_size,
i * input_size : (i + 1) * input_size,
]
if all(
isinstance(array_, ndarray) for arrays_ in arrays for array_ in arrays_
):
arrays[0][0] = csr_array(arrays[0][0])
diagonal_arrays.append(block_array(arrays))
self.__jacobian = block_diag(diagonal_arrays, format="csr")
def __initialize_jacobian(
self, jacobian_data: JacobianData, n_samples: int
) -> None:
"""Initialize the attribute ``__jacobian``.
Args:
jacobian_data: The discipline's Jacobian data.
n_samples: The number of samples.
"""
input_name = self.differentiated_input_names_substitute[0]
n_columns = self.__differentiated_input_size * n_samples
n_rows = sum(
jacobian_data[output_name][input_name].shape[-2]
for output_name in self.output_names
)
shape = n_columns if n_rows == 1 else (n_rows, n_columns)
create_array = dok_array if n_samples > 1 else empty
self.__jacobian = create_array(shape)
def __create_input_names_to_slices(self) -> None:
"""Create the map from discipline input names to input vector slices.
Raises:
ValueError: When a discipline input has no default value.
"""
input_data = self.__discipline.io.get_input_data()
input_data.update(self.__discipline.io.input_grammar.defaults)
self.__input_names_to_sizes.update(
self.__discipline.io.input_grammar.data_converter.compute_names_to_sizes(
input_data.keys(), input_data
)
)
missing_names = (
set(self.input_names)
.difference(self.__input_names_to_sizes.keys())
.difference(input_data.keys())
)
if missing_names:
msg = (
f"The size of the input {','.join(missing_names)} cannot be guessed "
f"from the discipline {self.__discipline.name}, "
f"nor from its default inputs or from its local data."
)
raise ValueError(msg)
(
self.__input_names_to_slices,
self.__input_size,
) = self.__discipline.io.input_grammar.data_converter.compute_names_to_slices(
self.input_names, input_data, self.__input_names_to_sizes
)
(
self.__differentiated_input_names_to_slices,
self.__differentiated_input_size,
) = self.__discipline.io.input_grammar.data_converter.compute_names_to_slices(
self.differentiated_input_names_substitute,
input_data,
self.__input_names_to_sizes,
)
def __create_discipline_input_data(
self,
x_vect: ndarray,
) -> dict[str, ndarray]:
"""Create the discipline input data from the function input vector.
The variables in the input data are cast according to the types defined in the
design space.
Args:
x_vect: The input vector of the function.
Returns:
The input data of the discipline.
"""
if self.__default_inputs:
self.__discipline.io.input_grammar.defaults.update(self.__default_inputs)
if not self.__input_names_to_slices:
self.__create_input_names_to_slices()
input_data = (
self.__discipline.io.input_grammar.data_converter.convert_array_to_data(
x_vect, self.__input_names_to_slices
)
)
variable_types = x_vect.dtype.metadata
if variable_types is not None:
# Restore the proper data types as declared in the design space.
for name, type_ in variable_types.items():
input_data[name] = input_data[name].astype(type_, copy=False)
if x_vect.ndim == 2:
return {k: v.ravel() for k, v in input_data.items()}
return input_data
