Source code for gemseo.utils.derivatives.derivatives_approx

# 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 - API and implementation and/or documentation
#       :author : Francois Gallard
#    OTHER AUTHORS   - MACROSCOPIC CHANGES
"""Finite differences approximation."""
from __future__ import annotations

import logging
import pickle
from multiprocessing import cpu_count
from numbers import Number
from typing import Iterable
from typing import Mapping
from typing import Sequence
from typing import Sized
from typing import TYPE_CHECKING

from gemseo.utils.derivatives.gradient_approximator import GradientApproximationFactory

if TYPE_CHECKING:
    from gemseo.core.discipline import MDODiscipline

from matplotlib import pyplot
from matplotlib.pyplot import Figure
from numpy import (
    absolute,
    allclose,
    amax,
    arange,
    array,
    atleast_2d,
    concatenate,
    divide,
    finfo,
    ndarray,
    zeros,
)

from gemseo.utils.data_conversion import (
    concatenate_dict_of_arrays_to_array,
    split_array_to_dict_of_arrays,
)
from pathlib import Path

EPSILON = finfo(float).eps
LOGGER = logging.getLogger(__name__)


[docs]class DisciplineJacApprox: """Approximates a discipline Jacobian using finite differences or Complex step.""" COMPLEX_STEP = "complex_step" FINITE_DIFFERENCES = "finite_differences" N_CPUS = cpu_count() def __init__( self, discipline: MDODiscipline, approx_method: str = FINITE_DIFFERENCES, step: Number | Iterable[Number] = 1e-7, parallel: bool = False, n_processes: int = N_CPUS, use_threading: bool = False, wait_time_between_fork: float = 0, ): """ Args: discipline: The discipline for which the Jacobian approximation shall be made. approx_method: The approximation method, either ``complex_step`` or ``finite_differences``. step: The differentiation step. The ``finite_differences`` takes either a float or an iterable of floats with the same length as the inputs. The ``complex_step`` method takes either a complex or a float as input. parallel: Whether to differentiate the discipline in parallel. n_processes: The maximum simultaneous number of threads, if ``use_threading`` is True, or processes otherwise, used to parallelize the execution. use_threading: Whether to use threads instead of processes to parallelize the execution; multiprocessing will copy (serialize) all the disciplines, while threading will share all the memory This is important to note if you want to execute the same discipline multiple times, you shall use multiprocessing. wait_time_between_fork: The time waited between two forks of the process / thread. """ from gemseo.core.mdofunctions.function_generator import MDOFunctionGenerator self.discipline = discipline self.approx_method = approx_method self.step = step self.generator = MDOFunctionGenerator(discipline) self.func = None self.approximator = None self.auto_steps = {} self.__par_args = { "n_processes": n_processes, "use_threading": use_threading, "wait_time_between_fork": wait_time_between_fork, } self.__parallel = parallel def _create_approximator( self, outputs: Sequence[str], inputs: Sequence[str], ): """Create the Jacobian approximation class. Args: inputs: The names of the inputs used to differentiate the outputs. outputs: The names of the outputs to be differentiated. Raises: ValueError: If the Jacobian approximation method is unknown. """ self.func = self.generator.get_function( input_names=inputs, output_names=outputs ) if self.approx_method not in [self.FINITE_DIFFERENCES, self.COMPLEX_STEP]: raise ValueError( f"Unknown Jacobian approximation method {self.approx_method}." ) factory = GradientApproximationFactory() self.approximator = factory.create( self.approx_method, self.func, step=self.step, parallel=self.__parallel, **self.__par_args, )
[docs] def auto_set_step( self, outputs: Sequence[str], inputs: Sequence[str], print_errors: bool = True, numerical_error: float = EPSILON, ) -> ndarray: r"""Compute the optimal step. Require a first evaluation of the perturbed functions values. The optimal step is reached when the truncation error (cut in the Taylor development), and the numerical cancellation errors (round-off when doing :math:`f(x+step)-f(x))` are equal. See Also: https://en.wikipedia.org/wiki/Numerical_differentiation and *Numerical Algorithms and Digital Representation*, Knut Morken, Chapter 11, "Numerical Differentiation" Args: inputs: The names of the inputs used to differentiate the outputs. outputs: The names of the outputs to be differentiated. print_errors: Whether to log the cancellation and truncation error estimates. numerical_error: The numerical error associated to the calculation of :math:`f`. By default Machine epsilon (appx 1e-16), but can be higher. when the calculation of :math:`f` requires a numerical resolution. Returns: The Jacobian of the function. """ self._create_approximator(outputs, inputs) old_cache_tol = self.discipline.cache_tol self.discipline.cache_tol = 0.0 x_vect = self._prepare_xvect(inputs, self.discipline.default_inputs) compute_opt_step = self.approximator.compute_optimal_step steps_opt, errors = compute_opt_step(x_vect, numerical_error=numerical_error) if print_errors: LOGGER.info( "Set optimal step for finite differences. " "Estimated approximation errors =" ) LOGGER.info(errors) self.discipline.cache_tol = old_cache_tol data = self.discipline.local_data data_sizes = {key: val.size for key, val in data.items()} self.auto_steps = split_array_to_dict_of_arrays(steps_opt, data_sizes, inputs) return errors, self.auto_steps
def _prepare_xvect( self, inputs: Iterable[str], data: dict[str, ndarray] | None = None, ) -> ndarray: """Convert an input data mapping into an input array. Args: inputs: The names of the inputs to be used for the differentiation. data: The input data mapping. If None, use the local data of the discipline. Returns: The input array. """ if data is None: data = self.discipline.local_data return concatenate_dict_of_arrays_to_array(data, inputs)
[docs] def compute_approx_jac( self, outputs: Iterable[str], inputs: Iterable[str], x_indices: Sequence[int] | None = None, ) -> dict[str, dict[str, ndarray]]: """Approximate the Jacobian. Args: outputs: The names of the outputs to be differentiated. inputs: The names of the inputs used to differentiate the outputs. x_indices: The components of the input vector to be used for the differentiation. If None, use all the components. Returns: The approximated Jacobian. """ self._create_approximator(outputs, inputs) old_cache_tol = self.discipline.cache_tol self.discipline.cache_tol = 0.0 local_data = self.discipline.local_data x_vect = self._prepare_xvect(inputs) if ( self.auto_steps is not None and array([key in self.auto_steps for key in inputs]).all() ): step = concatenate([self.auto_steps[key] for key in inputs]) else: step = self.step if isinstance(step, Sized) and 1 < len(step) != len(x_vect): raise ValueError( f"Inconsistent step size, expected {x_vect.size} got {len(step)}." ) flat_jac = self.approximator.f_gradient(x_vect, x_indices=x_indices, step=step) flat_jac = atleast_2d(flat_jac) data_sizes = {key: len(local_data[key]) for key in outputs} outputs_len = sum(data_sizes.values()) input_sizes = {key: len(local_data[key]) for key in inputs} inputs_len = sum(input_sizes.values()) data_sizes.update(input_sizes) global_shape = [outputs_len, inputs_len] if x_indices is None: flat_jac_complete = flat_jac else: flat_jac_complete = zeros(global_shape) flat_jac_complete[:, x_indices] = flat_jac self.discipline.cache_tol = old_cache_tol return split_array_to_dict_of_arrays( flat_jac_complete, data_sizes, outputs, inputs )
[docs] def check_jacobian( self, analytic_jacobian: dict[str, dict[str, ndarray]], outputs: Iterable[str], inputs: Iterable[str], discipline: MDODiscipline, threshold: float = 1e-8, plot_result: bool = False, file_path: str | Path = "jacobian_errors.pdf", show: bool = False, fig_size_x: float = 10.0, fig_size_y: float = 10.0, reference_jacobian_path: str | Path | None = None, save_reference_jacobian: bool = False, indices: int | Sequence[int] | slice | Ellipsis | None = None, ) -> bool: """Check if the analytical Jacobian is correct with respect to a reference one. If `reference_jacobian_path` is not `None` and `save_reference_jacobian` is `True`, compute the reference Jacobian with the approximation method and save it in `reference_jacobian_path`. If `reference_jacobian_path` is not `None` and `save_reference_jacobian` is `False`, do not compute the reference Jacobian but read it from `reference_jacobian_path`. If `reference_jacobian_path` is `None`, compute the reference Jacobian without saving it. Args: analytic_jacobian: The Jacobian to validate. inputs: The names of the inputs used to differentiate the outputs. outputs: The names of the outputs to be differentiated. discipline: The discipline to be differentiated. threshold: The acceptance threshold for the Jacobian error. plot_result: Whether to plot the result of the validation (computed vs approximated Jacobians). file_path: The path to the output file if ``plot_result`` is ``True``. show: Whether to open the figure. fig_size_x: The x-size of the figure in inches. fig_size_y: The y-size of the figure in inches. reference_jacobian_path: The path of the reference Jacobian file. save_reference_jacobian: Whether to save the reference Jacobian. indices: The indices of the inputs and outputs for the different sub-Jacobian matrices, formatted as ``{variable_name: variable_components}`` where ``variable_components`` can be either an integer, e.g. `2` a sequence of integers, e.g. `[0, 3]`, a slice, e.g. `slice(0,3)`, the ellipsis symbol (`...`) or `None`, which is the same as ellipsis. If a variable name is missing, consider all its components. If None, consider all the components of all the ``inputs`` and ``outputs``. Returns: Whether the analytical Jacobian is correct. """ inputs_indices = input_indices = outputs_indices = None if indices is not None: input_indices, inputs_indices = self._compute_variables_indices( indices, inputs, {name: len(self.discipline.default_inputs[name]) for name in inputs}, ) if reference_jacobian_path is None or save_reference_jacobian: approximated_jacobian = self.compute_approx_jac( outputs, inputs, input_indices ) else: with Path(reference_jacobian_path).open("rb") as infile: approximated_jacobian = pickle.load(infile) if save_reference_jacobian: with Path(reference_jacobian_path).open("wb") as outfile: pickle.dump(approximated_jacobian, outfile) name = discipline.name succeed = True if indices is not None: outputs_sizes = { output_name: output_jacobian[next(iter(output_jacobian))].shape[0] for output_name, output_jacobian in approximated_jacobian.items() } output_indices, outputs_indices = self._compute_variables_indices( indices, outputs, outputs_sizes ) if inputs_indices is None: inputs_indices = Ellipsis if outputs_indices is None: outputs_indices = Ellipsis for output_name, output_jacobian in approximated_jacobian.items(): for input_name, approx_jac in output_jacobian.items(): computed_jac = analytic_jacobian[output_name][input_name] if indices is not None: row_idx = atleast_2d(outputs_indices[output_name]).T col_idx = inputs_indices[input_name] computed_jac = computed_jac[row_idx, col_idx] approx_jac = approx_jac[row_idx, col_idx] if approx_jac.shape != computed_jac.shape: succeed = False msg = ( "{} Jacobian: dp {}/dp {} is of wrong shape; " "got: {} while expected: {}.".format( name, output_name, input_name, computed_jac.shape, approx_jac.shape, ) ) LOGGER.error(msg) else: success_loc = allclose( computed_jac, approx_jac, atol=threshold, rtol=threshold ) if not success_loc: err = amax( divide( absolute(computed_jac - approx_jac), absolute(approx_jac) + 1.0, ) ) LOGGER.error( "%s Jacobian: dp %s/d %s is wrong by %s%%.", name, output_name, input_name, err * 100.0, ) LOGGER.info("Approximate jacobian = \n%s", approx_jac) LOGGER.info("Provided by linearize method = \n%s", computed_jac) LOGGER.info( "Difference of jacobians = \n%s", approx_jac - computed_jac ) succeed = succeed and success_loc else: LOGGER.info( "Jacobian: dp %s/dp %s succeeded.", output_name, input_name ) LOGGER.info( "Linearization of MDODiscipline: %s is %s.", name, "correct" if succeed else "wrong", ) if plot_result: self.plot_jac_errors( analytic_jacobian, approximated_jacobian, file_path, show, fig_size_x, fig_size_y, ) return succeed
@staticmethod def _compute_variables_indices( indices: Mapping[str, int | Sequence[int] | Ellipsis | slice], variables_names: Iterable[str], variables_sizes: Mapping[str, int], ) -> list[int]: """Return indices. Args: indices: The indices for variables formatted as ``{variable_name: variable_components}`` where ``variable_components`` can be either an integer, e.g. `2` a sequence of integers, e.g. `[0, 3]`, a slice, e.g. `slice(0,3)`, the ellipsis symbol (`...`) or `None`, which is the same as ellipsis. If a variable name is missing, consider all its components. variables_names: The names of the variables. Returns: The indices of the variables. """ indices_sequence = [] variables_indices = {} variable_position = 0 for variable_name in variables_names: variable_size = variables_sizes[variable_name] variable_indices = list(range(variable_size)) indices_sequence.append(indices.get(variable_name, variable_indices)) if isinstance(indices_sequence[-1], int): indices_sequence[-1] = [indices_sequence[-1]] if isinstance(indices_sequence[-1], slice): indices_sequence[-1] = variable_indices[indices_sequence[-1]] if indices_sequence[-1] in [Ellipsis, None]: indices_sequence[-1] = variable_indices variables_indices[variable_name] = indices_sequence[-1] indices_sequence[-1] = [ variable_index + variable_position for variable_index in indices_sequence[-1] ] variable_position += variable_size indices_sequence = [item for sublist in indices_sequence for item in sublist] return indices_sequence, variables_indices @staticmethod def __concatenate_jacobian_per_output_names( analytic_jacobian: dict[str, dict[str, ndarray]], approximated_jacobian: dict[str, dict[str, ndarray]], ) -> tuple[dict[str, ndarray], dict[str, ndarray], list[str]]: """Concatenate the Jacobian matrices per output name. Args: analytic_jacobian: The reference Jacobian of the form ``{output_name: {input_name: sub_jacobian}}``. approximated_jacobian: The approximated Jacobian of the form ``{output_name: {input_name: sub_jacobian}}``. Returns: The analytic Jacobian of the form ``{output_name: sub_jacobian}``, the approximated Jacobian of the form ``{output_name: sub_jacobian}`` and the names of the output components corresponding to the columns of ``sub_jacobian``. """ _approx_jacobian = {} _analytic_jacobian = {} jacobian = analytic_jacobian[next(iter(analytic_jacobian))] input_names = list(jacobian.keys()) input_component_names = [ f"{input_name}_{i+1}" for input_name in input_names for i in range(jacobian[input_name].shape[1]) ] for output_name, output_approximated_jacobian in approximated_jacobian.items(): _output_approx_jacobian = concatenate( [ output_approximated_jacobian[input_name] for input_name in input_names ], axis=1, ) _output_analytic_jacobian = concatenate( [ analytic_jacobian[output_name][input_name] for input_name in input_names ], axis=1, ) n_f, _ = _output_approx_jacobian.shape if n_f == 1: _approx_jacobian[output_name] = _output_approx_jacobian.flatten() _analytic_jacobian[output_name] = _output_analytic_jacobian.flatten() else: for i in range(n_f): output_name = f"{output_name}_{i}" _approx_jacobian[output_name] = _output_approx_jacobian[i, :] _analytic_jacobian[output_name] = _output_analytic_jacobian[i, :] return _analytic_jacobian, _approx_jacobian, input_component_names
[docs] def plot_jac_errors( self, computed_jac: ndarray, approx_jac: ndarray, file_path: str | Path = "jacobian_errors.pdf", show: bool = False, fig_size_x: float = 10.0, fig_size_y: float = 10.0, ) -> Figure: """Generate a plot of the exact vs approximated Jacobian. Args: computed_jac: The Jacobian to validate. approx_jac: The approximated Jacobian. file_path: The path to the output file if ``plot_result`` is ``True``. show: Whether to open the figure. fig_size_x: The x-size of the figure in inches. fig_size_y: The y-size of the figure in inches. """ comp_grad, app_grad, x_labels = self.__concatenate_jacobian_per_output_names( computed_jac, approx_jac ) n_funcs = len(app_grad) if n_funcs == 0: raise ValueError("No gradients to plot!") nrows = n_funcs // 2 if 2 * nrows < n_funcs: nrows += 1 ncols = 2 fig, axes = pyplot.subplots( nrows=nrows, ncols=2, sharex=True, sharey=False, figsize=(fig_size_x, fig_size_y), ) i = 0 j = -1 axes = atleast_2d(axes) n_subplots = len(axes) * len(axes[0]) abscissa = arange(len(x_labels)) for func, grad in sorted(comp_grad.items()): j += 1 if j == ncols: j = 0 i += 1 axe = axes[i][j] axe.plot(abscissa, grad, "bo") axe.plot(abscissa, app_grad[func], "ro") axe.set_title(func) axe.set_xticklabels(x_labels, fontsize=14) axe.set_xticks(abscissa) for tick in axe.get_xticklabels(): tick.set_rotation(90) # Update y labels spacing vis_labels = [ label for label in axe.get_yticklabels() if label.get_visible() is True ] pyplot.setp(vis_labels[::2], visible=False) # pyplot.xticks(rotation=90) if len(comp_grad.items()) < n_subplots: # xlabel must be written with the same fontsize on the 2 columns j += 1 axe = axes[i][j] axe.set_xticklabels(x_labels, fontsize=14) axe.set_xticks(abscissa) for tick in axe.get_xticklabels(): tick.set_rotation(90) fig.suptitle( "Computed and approximate derivatives. " + " blue = computed, red = approximated derivatives", fontsize=14, ) if file_path is not None: pyplot.savefig(file_path) if show: pyplot.show() return fig
[docs]def comp_best_step( f_p: ndarray, f_x: ndarray, f_m: ndarray, step: float, epsilon_mach: float = EPSILON, ) -> tuple[ndarray | None, ndarray | None, float]: r"""Compute the optimal step for finite differentiation. Applied to a forward first order finite differences gradient approximation. Require a first evaluation of the perturbed functions values. The optimal step is reached when the truncation error (cut in the Taylor development), and the numerical cancellation errors (round-off when doing :math:`f(x+step)-f(x))` are equal. See Also: https://en.wikipedia.org/wiki/Numerical_differentiation and *Numerical Algorithms and Digital Representation*, Knut Morken, Chapter 11, "Numerical Differenciation" Args: f_p: The value of the function :math:`f` at the next step :math:`x+\\delta_x`. f_x: The value of the function :math:`f` at the current step :math:`x`. f_m: The value of the function :math:`f` at the previous step :math:`x-\\delta_x`. step: The differentiation step :math:`\\delta_x`. Returns: The estimation of the truncation error. None if the Hessian approximation is too small to compute the optimal step. The estimation of the cancellation error. None if the Hessian approximation is too small to compute the optimal step. The optimal step. """ hess = approx_hess(f_p, f_x, f_m, step) if abs(hess) < 1e-10: LOGGER.debug("Hessian approximation is too small, can't compute optimal step.") return None, None, step opt_step = 2 * (epsilon_mach * abs(f_x) / abs(hess)) ** 0.5 trunc_error = compute_truncature_error(hess, step) cancel_error = compute_cancellation_error(f_x, opt_step) return trunc_error, cancel_error, opt_step
[docs]def compute_truncature_error( hess: ndarray, step: float, ) -> ndarray: r"""Estimate the truncation error. Defined for a first order finite differences scheme. Args: hess: The second-order derivative :math:`d^2f/dx^2`. step: The differentiation step. Returns: The truncation error. """ return abs(hess) * step / 2
[docs]def compute_cancellation_error( f_x: ndarray, step: float, epsilon_mach=EPSILON, ) -> ndarray: r"""Estimate the cancellation error. This is the round-off when doing :math:`f(x+\\delta_x)-f(x)`. Args: f_x: The value of the function at the current step :math:`x`. step: The step used for the calculations of the perturbed functions values. epsilon_mach: The machine epsilon. Returns: The cancellation error. """ return 2 * epsilon_mach * abs(f_x) / step
[docs]def approx_hess( f_p: ndarray, f_x: ndarray, f_m: ndarray, step: float, ) -> ndarray: r"""Compute the second-order approximation of the Hessian matrix :math:`d^2f/dx^2`. Args: f_p: The value of the function :math:`f` at the next step :math:`x+\\delta_x`. f_x: The value of the function :math:`f` at the current step :math:`x`. f_m: The value of the function :math:`f` at the previous step :math:`x-\\delta_x`. step: The differentiation step :math:`\\delta_x`. Returns: The approximation of the Hessian matrix at the current step :math:`x`. """ return (f_p - 2 * f_x + f_m) / (step**2)