# 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: Matthias De Lozzo
# OTHER AUTHORS - MACROSCOPIC CHANGES
"""Surrogate discipline."""
from __future__ import annotations
import logging
from typing import Iterable
from typing import Mapping
from numpy import ndarray
from gemseo.core.dataset import Dataset
from gemseo.core.discipline import MDODiscipline
from gemseo.core.jacobian_assembly import JacobianAssembly
from gemseo.mlearning.core.ml_algo import MLAlgoParameterType
from gemseo.mlearning.core.ml_algo import TransformerType
from gemseo.mlearning.regression.factory import RegressionModelFactory
from gemseo.mlearning.regression.regression import MLRegressionAlgo
from gemseo.utils.string_tools import MultiLineString
from gemseo.utils.string_tools import pretty_repr
LOGGER = logging.getLogger(__name__)
[docs]class SurrogateDiscipline(MDODiscipline):
"""A :class:`.MDODiscipline` approximating another one with a surrogate model.
This surrogate model is a regression model implemented as a
:class:`.MLRegressionAlgo`. This :class:`.MLRegressionAlgo` is built from an input-
output :class:`.Dataset` composed of evaluations of the original discipline.
"""
_ATTR_TO_SERIALIZE = MDODiscipline._ATTR_TO_SERIALIZE + ("regression_model",)
def __init__(
self,
surrogate: str | MLRegressionAlgo,
data: Dataset | None = None,
transformer: TransformerType | None = MLRegressionAlgo.DEFAULT_TRANSFORMER,
disc_name: str | None = None,
default_inputs: dict[str, ndarray] | None = None,
input_names: Iterable[str] | None = None,
output_names: Iterable[str] | None = None,
**parameters: MLAlgoParameterType,
) -> None:
""".. # noqa: D205 D212 D415
Args:
surrogate: Either the class name
or the instance of the :class:`.MLRegressionAlgo`.
data: The learning dataset to train the regression model.
If None, the regression model is supposed to be trained.
transformer: The strategies to transform the variables.
The values are instances of :class:`.Transformer`
while the keys are the names of
either the variables
or the groups of variables,
e.g. "inputs" or "outputs" in the case of the regression algorithms.
If a group is specified,
the :class:`.Transformer` will be applied
to all the variables of this group.
If None, do not transform the variables.
The :attr:`.MLRegressionAlgo.DEFAULT_TRANSFORMER` uses
the :class:`.MinMaxScaler` strategy for both input and output variables.
disc_name: The name to be given to the surrogate discipline.
If None, concatenate :attr:`.SHORT_ALGO_NAME` and ``data.name``.
default_inputs: The default values of the inputs.
If None, use the center of the learning input space.
input_names: The names of the input variables.
If None, consider all input variables mentioned in the learning dataset.
output_names: The names of the output variables.
If None, consider all input variables mentioned in the learning dataset.
**parameters: The parameters of the machine learning algorithm.
Raises:
ValueError: If the learning dataset is missing
whilst the regression model is not trained.
"""
if isinstance(surrogate, MLRegressionAlgo):
self.regression_model = surrogate
name = self.regression_model.learning_set.name
elif data is None:
raise ValueError("data is required to train the surrogate model.")
else:
factory = RegressionModelFactory()
self.regression_model = factory.create(
surrogate,
data=data,
transformer=transformer,
input_names=input_names,
output_names=output_names,
**parameters,
)
name = f"{self.regression_model.SHORT_ALGO_NAME}_{data.name}"
disc_name = disc_name or name
if not self.regression_model.is_trained:
self.regression_model.learn()
msg = MultiLineString()
msg.add("Build the surrogate discipline: {}", disc_name)
msg.indent()
msg.add("Dataset name: {}", data.name)
msg.add("Dataset size: {}", data.length)
msg.add("Surrogate model: {}", self.regression_model.__class__.__name__)
LOGGER.info("%s", msg)
if not name.startswith(self.regression_model.SHORT_ALGO_NAME):
disc_name = f"{self.regression_model.SHORT_ALGO_NAME}_{disc_name}"
msg = MultiLineString()
msg.add("Use the surrogate discipline: {}", disc_name)
msg.indent()
super().__init__(disc_name)
self._initialize_grammars(input_names, output_names)
msg.add("Inputs: {}", pretty_repr(self.get_input_data_names()))
msg.add("Outputs: {}", pretty_repr(self.get_output_data_names()))
self._set_default_inputs(default_inputs)
self.add_differentiated_inputs()
self.add_differentiated_outputs()
try:
self.regression_model.predict_jacobian(self.default_inputs)
self.linearization_mode = JacobianAssembly.AUTO_MODE
msg.add("Jacobian: use surrogate model jacobian")
except NotImplementedError:
self.linearization_mode = self.FINITE_DIFFERENCES
msg.add("Jacobian: use finite differences")
LOGGER.info("%s", msg)
def __repr__(self) -> str:
model = self.regression_model.__class__.__name__
data_name = self.regression_model.learning_set.name
length = len(self.regression_model.learning_set)
inputs = sorted(self.regression_model.input_names)
outputs = sorted(self.regression_model.output_names)
arguments = [
f"name={self.name}",
f"algo={model}",
f"data={data_name}",
f"size={length}",
f"inputs=[{pretty_repr(inputs)}]",
f"outputs=[{pretty_repr(outputs)}]",
f"jacobian={self.linearization_mode}",
]
msg = "SurrogateDiscipline({})".format(", ".join(arguments))
return msg
def __str__(self) -> str:
data_name = self.regression_model.learning_set.name
length = len(self.regression_model.learning_set)
msg = MultiLineString()
msg.add("Surrogate discipline: {}", self.name)
msg.indent()
msg.add("Dataset name: {}", data_name)
msg.add("Dataset size: {}", length)
msg.add("Surrogate model: {}", self.regression_model.__class__.__name__)
inputs = sorted(self.regression_model.input_names)
outputs = sorted(self.regression_model.output_names)
msg.add("Inputs: {}", pretty_repr(inputs))
msg.add("Outputs: {}", pretty_repr(outputs))
return str(msg)
def _initialize_grammars(
self,
input_names: Iterable[str] | None = None,
output_names: Iterable[str] | None = None,
) -> None:
"""Initialize the input and output grammars from the regression model.
Args:
input_names: The names of the inputs to consider.
If None, use all the inputs of the regression model.
output_names: The names of the inputs to consider.
If None, use all the inputs of the regression model.
"""
self.input_grammar.update(input_names or self.regression_model.input_names)
self.output_grammar.update(output_names or self.regression_model.output_names)
def _set_default_inputs(
self,
default_inputs: Mapping[str, ndarray] = None,
) -> None:
"""Set the default values of the inputs.
Args:
default_inputs: The default values of the inputs.
If None, use the the center of the learning input space.
"""
if default_inputs is None:
self.default_inputs = self.regression_model.input_space_center
else:
self.default_inputs = default_inputs
def _run(self) -> None:
input_data = self.get_input_data()
output_data = self.regression_model.predict(input_data)
output_data = {key: val.flatten() for key, val in output_data.items()}
self.local_data.update(output_data)
def _compute_jacobian(
self,
inputs: Iterable[str] | None = None,
outputs: Iterable[str] | None = None,
) -> None:
input_data = self.get_input_data()
self._init_jacobian(inputs, outputs)
self.jac = self.regression_model.predict_jacobian(input_data)