Source code for gemseo.problems.scalable.parametric.study

# -*- coding: utf-8 -*-
# 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
"""
Scalable study
==============
"""
from __future__ import division, unicode_literals

import logging
import os
import pickle

from matplotlib import pyplot as plt
from numpy import arange, where
from numpy.random import rand

from gemseo.core.coupling_structure import MDOCouplingStructure
from gemseo.core.mdo_scenario import MDOScenario
from gemseo.mda.mda_factory import MDAFactory
from gemseo.problems.scalable.parametric.core.variables import get_constraint_name
from gemseo.problems.scalable.parametric.problem import TMScalableProblem

LOGGER = logging.getLogger(__name__)

MDA_TOLERANCE = {"tolerance": 1e-14, "linear_solver_tolerance": 1e-14}
ALGO_OPTIONS = {
    "xtol_rel": 1e-4,
    "ftol_rel": 1e-4,
    "xtol_abs": 1e-4,
    "ftol_abs": 1e-4,
    "ineq_tolerance": 1e-3,
    "eq_tolerance": 1e-3,
}


[docs]def save_matrix_plot(matrix, disc, name, directory="."): """Save the graphical representation of a matrix. :param ndarray matrix: matrix. :param str disc: discipline name. :param str name: name of the matrix. :param str directory: name of the directory to write the file. Default: '.'. """ plt.matshow(matrix, cmap=plt.get_cmap("binary"), vmin=0.0, vmax=1.0) plt.colorbar() filename = disc + "_" + name plt.savefig(os.path.join(directory, filename + ".pdf")) plt.close()
OBJECTIVE_NAME = "obj" COUPLING_DIR = "coupling" COEFF_DIR = "coefficients" OPTIM_DIR = "opthistoryview"
[docs]class TMParamSS(object): """This scalable parametric study realizes scalable studies with different scaling strategies. E.g. comparison of MDF and IDF formulations in terms of execution time for different number of coupling variables. """ def __init__( self, n_disciplines, n_shared, n_local, n_coupling, full_coupling=True, active_probability=0.1, feasibility_level=0.8, seed=1, directory="results", ): """The TMParamSS constructor depends on: - the number of disciplines, - the number of shared design parameters, - the number of local design parameters for each discipline, - the number of coupling variables for each each discipline. One of the three latter should be a list of integers whose components define scaling strategies. The constructor instantiates as many TMScalableStudy as scaling strategies. :param int n_disciplines: number of disciplines. :param n_shared: (list of) number(s) of shared design parameters. :param n_local: (list of) number(s) of local design parameters for each discipline. :param n_coupling: (list of) number(s) of coupling parameters for each discipline. :param bool full_coupling: fully couple the disciplines, ie each TMDiscipline depends on each other. Default: True. :param float active_probability: active probability :param float feasibility_level: level of feasibility :param int seed: seed for replicability. :param str directory: directory to store results See also -------- .TMScalableStudy : standard scalable study launched with different configurations in the case of parametric scalable study. """ self.n_shared = n_shared self.n_local = n_local self.n_coupling = n_coupling self.n_disciplines = n_disciplines assert isinstance(n_disciplines, int) assert isinstance(n_shared, (int, list)) assert isinstance(n_local, (int, list)) assert isinstance(n_coupling, (int, list)) n_lists = sum( [ 1 if isinstance(val, list) else 0 for val in (n_shared, n_local, n_coupling) ] ) if n_lists > 1: msg = "At most 1 value among (n_shared,n_local,n_coupling) can be a list !" raise ValueError(msg) if isinstance(n_shared, list): self.param_label = "n_shared" self.param = n_shared self.studies = [ TMScalableStudy( n_disciplines, n_shared_i, n_local, n_coupling, full_coupling, active_probability, feasibility_level, seed=seed, directory=directory, ) for n_shared_i in n_shared ] elif isinstance(n_local, list): self.param_label = "n_local" self.param = n_local self.studies = [ TMScalableStudy( n_disciplines, n_shared, n_local_i, n_coupling, full_coupling, active_probability, feasibility_level, seed=seed, directory=directory, ) for n_local_i in n_local ] elif isinstance(n_coupling, list): self.param_label = "n_coupling" self.param = n_coupling self.studies = [ TMScalableStudy( n_disciplines, n_shared, n_local, n_coupling_i, full_coupling, active_probability, feasibility_level, seed, directory=directory, ) for n_coupling_i in n_coupling ] else: self.param_label = "None" self.param = [1] self.studies = [ TMScalableStudy( n_disciplines, n_shared, n_local, n_coupling, full_coupling, active_probability, feasibility_level, seed, directory=directory, ) ] def __str__(self): msg = "Parametric scalable study\n" msg += "> " + str(self.n_disciplines) + " disciplines\n" if hasattr(self.n_shared, "__len__"): n_shared = [str(val) for val in self.n_shared] n_shared = ", ".join(n_shared) tmp = n_shared.split(", ") n_shared = ", ".join(tmp[0:-1]) + " or " + tmp[-1] else: n_shared = str(self.n_shared) if hasattr(self.n_local, "__len__"): n_local = [str(val) for val in self.n_local] n_local = ", ".join(n_local) tmp = n_local.split(", ") n_local = ", ".join(tmp[0:-1]) + " or " + tmp[-1] else: n_local = str(self.n_local) if hasattr(self.n_coupling, "__len__"): n_coupling = [str(val) for val in self.n_coupling] n_coupling = ", ".join(n_coupling) tmp = n_coupling.split(", ") n_coupling = ", ".join(tmp[0:-1]) + " or " + tmp[-1] else: n_coupling = str(self.n_coupling) msg += "> " + str(n_shared) + " shared design parameters\n" msg += "> " + str(n_local) msg += " local design parameters per discipline\n" msg += "> " + str(n_coupling) msg += " coupling variables per discipline\n" return msg
[docs] def run_formulation( self, formulation, max_iter=100, post_coupling=True, post_optim=True, post_coeff=True, algo="NLOPT_SLSQP", algo_options=None, ): """This method solves the scalable problems with a particular MDO formulation. :param str formulation: MDO formulation name :param int max_iter: maximum number of iterations :param bool post_coupling: store coupling plots :param bool post_optim: store optimization plots :param bool post_coeff: store coefficients plots :param algo: algorithm name to solve the problem :param algo_options: inequality and equality tolerance, xtol etc.. """ if algo_options is None: algo_options = ALGO_OPTIONS for study in self.studies: study.run_formulation( formulation, max_iter, post_coupling, post_optim, post_coeff, algo, algo_options, )
[docs] def save(self, file_path): """This method saves the results into a pickle file. :param str file_path: pickle file path to store the results. """ exec_time = {} sizes = [] for idx, _ in enumerate(self.studies): if self.param_label in ["n_local", "n_coupling"]: pos = self.param[idx] * self.studies[idx].n_disciplines else: pos = self.param[idx] exec_time[pos] = {} for formulation in self.studies[idx].formulations: e_t = self.studies[idx].exec_time[formulation] exec_time[pos][formulation] = e_t["scenario"] sizes.append(pos) with open(file_path, "wb") as f_out: pickle.dump( { "exec_time": exec_time, "sizes": sizes, "scaling": self.param_label, "formulations": self.studies[0].formulations, }, f_out, )
[docs]class TMParamSSPost(object): """This class is dedicated to the post-treatment of TMParamSS results.""" def __init__(self, file_path): """The constructor reads data stored in a pickle file. :param str file_path: file path where data are stored. """ with open(file_path, "rb") as f_in: results = pickle.load(f_in) self.exec_time = results["exec_time"] self.sizes = results["sizes"] self.scaling = results["scaling"] self.formulations = results["formulations"]
[docs] def plot( self, title="A scalable comparison of MDO formulations", save=False, show=True, file_path="comparison.pdf", ): """Plot one line per MDO formulation where the y-axis represents the execution time and the x-axis the scaling strategies. :param str title: title of the figure. Default: 'A scalable comparison of MDO formulations'. :param bool save: save the plot. Default: False. :param bool show: show the plot. Default: True. :param str file_path: file path to store the figure. Default: 'comparison.pdf'. """ _, axis = plt.subplots() colors = ["b", "r", "g", "c", "m", "y", "k", "w"] idx = 0 for formulation in self.formulations: exec_time = [self.exec_time[size][formulation] for size in self.sizes] axis.plot(self.sizes, exec_time, color=colors[idx], lw=2, label=formulation) idx += 1 axis.set_xscale("log") axis.set_yscale("log") if self.scaling == "n_local": axis.set_xlabel("Number of Local Design Variables") elif self.scaling == "n_coupling": axis.set_xlabel("Number of Coupling Variables") elif self.scaling == "n_shared": axis.set_xlabel("Number of Shared Design Variables") else: axis.set_xlabel("Scalable strategy") axis.set_ylabel("Optimization Time (s)") axis.set_title(title) axis.set_axisbelow(True) axis.grid(which="both", lw=1, color="gray") axis.legend() if save: plt.savefig(file_path) if show: plt.show()
[docs]class TMScalableStudy(object): """This scalable study creates a scalable MDO problem from Tedford and Martins, 2010 and compares its resolution according to different MDO formulations.""" def __init__( self, n_disciplines, n_shared, n_local, n_coupling, full_coupling=True, active_probability=0.1, feasibility_level=0.8, seed=1, directory="results", ): """The TMScalableStudy constructor depends on: - the number of disciplines, - the number of shared design parameters, - the number of local design parameters for each discipline, - the number of coupling variables for each each discipline. :param int n_disciplines: number of disciplines. :param int n_shared: number of shared design parameters. :param int n_local: number of local design parameters for each discipline. :param int n_coupling: number of coupling parameters for each discipline. :param bool full_coupling: fully couple the disciplines, ie each TMDiscipline depends on each other. Default: True. :param float active_probability: active probability :param float feasibility_level: level of feasibility :param str directory: directory to store results :param int seed: seed for replicability. See also -------- .TMScalableProblem : Scalable problem managed by the scalable study and providing both disciplines and design space. """ self.n_disciplines = n_disciplines self.n_local = n_local self.n_shared = n_shared self.n_coupling = n_coupling n_local = [n_local] * n_disciplines n_coupling = [n_coupling] * n_disciplines self.problem = TMScalableProblem( n_shared, n_local, n_coupling, full_coupling, seed=seed ) self.n_calls = {} self.n_calls_linearize = {} self.exec_time = {} self.formulation_options = {"MDF": {"sub_mda_class": "MDAGaussSeidel"}} self.formulation_options["MDF"].update(MDA_TOLERANCE) self.disc_names = ["scenario", "mda", "mdo_chain", "sub_mda"] tmp = sorted([disc.name for disc in self.problem.disciplines]) self.disc_names += tmp self.active_probability = active_probability self.feasibility_level = feasibility_level self.directory = directory def __store_statistics(self, formulation, scenario): """Store statistics in dictionaries. :param str formulation: MDO formulation. :param Scenario scenario: scenario. """ self.n_calls[formulation] = {} self.n_calls_linearize[formulation] = {} self.exec_time[formulation] = {} for disc in self.problem.disciplines: self.n_calls[formulation][disc.name] = disc.n_calls ncl = disc.n_calls_linearize self.n_calls_linearize[formulation][disc.name] = ncl self.exec_time[formulation][disc.name] = disc.exec_time self.exec_time[formulation]["scenario"] = scenario.exec_time ncl = scenario.n_calls_linearize self.n_calls_linearize[formulation]["scenario"] = ncl self.n_calls[formulation]["scenario"] = scenario.n_calls if hasattr(scenario.formulation, "mda"): mda = scenario.formulation.mda mda_exec_time = mda.exec_time mda_nc = mda.n_calls mda_ncl = mda.n_calls_linearize mdo_chain_exec_time = mda.mdo_chain.exec_time mdo_chain_nc = mda.mdo_chain.n_calls mdo_chain_ncl = mda.mdo_chain.n_calls_linearize sub_mda_exec_time = mda.sub_mda_list[0].exec_time sub_mda_nc = mda.sub_mda_list[0].n_calls sub_mda_ncl = mda.sub_mda_list[0].n_calls_linearize else: mda_exec_time = 0.0 mda_nc = 0 mda_ncl = 0 mdo_chain_exec_time = 0.0 mdo_chain_nc = 0 mdo_chain_ncl = 0 sub_mda_exec_time = 0.0 sub_mda_nc = 0 sub_mda_ncl = 0 self.exec_time[formulation]["mda"] = mda_exec_time self.exec_time[formulation]["sub_mda"] = sub_mda_exec_time self.exec_time[formulation]["mdo_chain"] = mdo_chain_exec_time self.n_calls[formulation]["mda"] = mda_nc self.n_calls[formulation]["sub_mda"] = sub_mda_nc self.n_calls[formulation]["mdo_chain"] = mdo_chain_nc self.n_calls_linearize[formulation]["mda"] = mda_ncl self.n_calls_linearize[formulation]["sub_mda"] = sub_mda_ncl self.n_calls_linearize[formulation]["mdo_chain"] = mdo_chain_ncl
[docs] def run_formulation( self, formulation, max_iter=100, post_coupling=True, post_optim=True, post_coeff=True, algo="NLOPT_SLSQP", algo_options=None, ): """This method solves the scalable problem with a particular MDO formulation. :param str formulation: MDO formulation name :param int max_iter: maximum number of iterations :param bool post_coupling: store coupling plots :param bool post_optim: store optimization plots :param bool post_coeff: store coefficients plots :param algo: algorithm name to solve the problem :param algo_options: inequality and equality tolerance, xtol etc.. """ if algo_options is None: algo_options = ALGO_OPTIONS self.problem.reset_design_space() self.problem.reset_disciplines() factory = MDAFactory() mda = factory.create( "MDAGaussSeidel", self.problem.disciplines, **MDA_TOLERANCE ) equilibrium = mda.execute() scenario = MDOScenario( self.problem.disciplines, formulation, OBJECTIVE_NAME, self.problem.design_space, **self.formulation_options.get(formulation, {}) ) LOGGER.info("Make the starting point feasible.") for disc in range(self.n_disciplines): if get_constraint_name(disc) in list(equilibrium.keys()): urand = rand(len(equilibrium[get_constraint_name(disc)])) val = equilibrium[get_constraint_name(disc)] alt = self.feasibility_level alt += (1 - self.feasibility_level) * val tau = where(urand < self.active_probability, val, alt) else: tau = 0.0 scenario.add_constraint(get_constraint_name(disc), "ineq", value=tau) input_data = {"algo": algo, "max_iter": max_iter, "algo_options": algo_options} scenario.execute(input_data) if post_coeff: mkdir(self.directory) path = mkdir(self.directory, COEFF_DIR) for disc in self.problem.disciplines[1:]: save_matrix_plot(disc.model.c_shared, disc.name, "c_shared", path) save_matrix_plot(disc.model.c_local, disc.name, "c_local", path) for name, c_coupling_val in disc.model.c_coupling.items(): save_matrix_plot( c_coupling_val, disc.name, "c_coupling_" + name, path ) if post_coupling: path = mkdir(self.directory, COUPLING_DIR) scenario.xdsmize( latex_output=True, outdir=path, outfilename=formulation + "_xdsm" ) coupling_structure = MDOCouplingStructure(scenario.disciplines) coupling_structure.plot_n2_chart(file_path=os.path.join(path, "n2.pdf")) if post_optim: path = mkdir(self.directory, OPTIM_DIR) + "/" scenario.post_process( "OptHistoryView", save=True, show=False, file_path=path ) self.__store_statistics(formulation, scenario) n_iter = 0 for discipline in scenario.disciplines: n_iter += discipline.n_calls n_iter += discipline.n_calls_linearize return { "x_opt": scenario.optimization_result.x_opt, "f_opt": scenario.optimization_result.f_opt, "status": scenario.optimization_result.status, "n_iter": n_iter, "is_feas": scenario.optimization_result.is_feasible, "exec_time": scenario.exec_time, }
@property def formulations(self): """Names of the MDO formulations. :return: list of MDO formulations names :rtype: list(str) """ return list(self.n_calls.keys()) def __str__(self): """String representation of the results (number of calls, number of linearizations and execution time) for each discipline. :return: string representation :rtype: str """ msg = [ "Scalable study", ".... {} disciplines".format(self.n_disciplines), ".... {} shared design parameters".format(self.n_shared), ".... {} local design parameters per discipline".format(self.n_local), ".... {} coupling variables per discipline".format(self.n_coupling), ] if self.formulations: msg.append("MDO formulations") for formulation in self.formulations: msg.append(".... {}".format(formulation)) for discipline in self.problem.disciplines: msg.append(self.__elementary_str(formulation, discipline.name)) if "mda" in self.exec_time[formulation]: msg.append(self.__elementary_str(formulation, "mda")) if "mdo_chain" in self.exec_time[formulation]: msg.append(self.__elementary_str(formulation, "mdo_chain")) if "sub_mda" in self.exec_time[formulation]: msg.append(self.__elementary_str(formulation, "sub_mda")) msg.append(self.__elementary_str(formulation, "scenario")) return "\n".join(msg) def __elementary_str(self, formulation, discipline): """String representation of the result for a given formulation and a given discipline. :param str formulation: MDO formulation name :param str discipline: discipline name :return: elementary string representation :rtype: str """ n_calls = self.n_calls[formulation].get(discipline, "NA") n_lin = self.n_calls_linearize[formulation].get(discipline, "NA") exec_time = self.exec_time[formulation].get(discipline, "NA") msg = "........ {} = {} calls / {} linearizations / {} seconds" return msg.format(discipline, n_calls, n_lin, exec_time)
[docs] def plot_exec_time(self, show=True, save=False, file_path="exec_time.pdf"): """Barplot of the execution time of the different disciplines for the different formulations. When the formulation is based on a MDA, the MDO scenario is detailed in terms of MDA, MDO chain and sub-MDA. :param bool show: if True, show plot. Default: False. :param bool save: if True, save plot. Default: False. :parma str file_path: file path. Default: "exec_time.pdf" """ series = [ [self.exec_time[formulation][disc] for disc in self.disc_names] for formulation in self.formulations ] bar_width = 0.9 / len(self.formulations) indices = arange(len(self.exec_time[self.formulations[0]])) factor = 0 colors = ["b", "g", "r", "c", "m", "y", "k", "w"] for formulation in self.formulations: plt.bar( indices + bar_width * factor, series[factor], bar_width, alpha=0.8, color=colors[factor], label=formulation, ) for name in ["mda", "mdo_chain", "sub_mda"]: plt.bar( [0.0 + bar_width * factor], [self.exec_time[formulation][name]], bar_width, fill=False, ) factor += 1 plt.xlabel("Disciplines") plt.ylabel("Execution time") plt.xticks(indices + bar_width, self.disc_names) plt.legend() plt.tight_layout() if save: plt.savefig(file_path) if show: plt.show() plt.close()
[docs]def mkdir(dirname, subdirname=None): """Create a directory if not exists. :param str dirname: name of the directory. :param str subdirname: name of the sub-directory. If None, only considers the directory. Default: None. """ if subdirname is not None: dirpath = os.path.join(dirname, subdirname) else: dirpath = dirname if not os.path.exists(dirpath): os.mkdir(dirpath) return dirpath