Note

Go to the end to download the full example code.

MDF-based MDO on the Sobieski SSBJ test case.#

from __future__ import annotations

from gemseo import configure_logger
from gemseo import create_discipline
from gemseo import create_scenario
from gemseo import generate_n2_plot
from gemseo.problems.mdo.sobieski.core.design_space import SobieskiDesignSpace

configure_logger()

<RootLogger root (INFO)>

Instantiate the disciplines#

First, we instantiate the four disciplines of the use case: SobieskiPropulsion, SobieskiAerodynamics, SobieskiMission and SobieskiStructure.

disciplines = create_discipline([
    "SobieskiPropulsion",
    "SobieskiAerodynamics",
    "SobieskiMission",
    "SobieskiStructure",
])

We can quickly access the most relevant information of any discipline (name, inputs, and outputs) with Python's print() function. Moreover, we can get the default input values of a discipline with the attribute Discipline.default_input_data

for discipline in disciplines:
    print(discipline)  # noqa: T201
    print(f"Default inputs: {discipline.default_input_data}")  # noqa: T201

SobieskiPropulsion
Default inputs: {'y_23': array([12562.01206488]), 'x_3': array([0.5]), 'x_shared': array([5.0e-02, 4.5e+04, 1.6e+00, 5.5e+00, 5.5e+01, 1.0e+03]), 'c_3': array([4360.])}
SobieskiAerodynamics
Default inputs: {'x_2': array([1.]), 'y_32': array([0.50279625]), 'x_shared': array([5.0e-02, 4.5e+04, 1.6e+00, 5.5e+00, 5.5e+01, 1.0e+03]), 'y_12': array([5.06069742e+04, 9.50000000e-01]), 'c_4': array([0.01375])}
SobieskiMission
Default inputs: {'y_14': array([50606.9741711 ,  7306.20262124]), 'x_shared': array([5.0e-02, 4.5e+04, 1.6e+00, 5.5e+00, 5.5e+01, 1.0e+03]), 'y_24': array([4.15006276]), 'y_34': array([1.10754577])}
SobieskiStructure
Default inputs: {'y_21': array([50606.9741711]), 'y_31': array([6354.32430691]), 'x_1': array([0.25, 1.  ]), 'x_shared': array([5.0e-02, 4.5e+04, 1.6e+00, 5.5e+00, 5.5e+01, 1.0e+03]), 'c_0': array([2000.]), 'c_1': array([25000.]), 'c_2': array([6.])}

You may also be interested in plotting the couplings of your disciplines. A quick way of getting this information is the high-level function generate_n2_plot(). A much more detailed explanation of coupling visualization is available here.

generate_n2_plot(disciplines, save=False, show=True)
plot sobieski mdf example

Build, execute and post-process the scenario#

Then, we build the scenario which links the disciplines with the formulation and the optimization algorithm. Here, we use the MDF formulation. We tell the scenario to minimize -y_4 instead of minimizing y_4 (range), which is the default option.

Instantiate the scenario#

During the instantiation of the scenario, we provide some options for the MDF formulations. The MDF formulation includes an MDA, and thus one of the settings of the formulation is main_mda_settings, which configures the solver for the strong couplings.

main_mda_settings = {
    "tolerance": 1e-14,
    "max_mda_iter": 50,
    "warm_start": True,
    "use_lu_fact": False,
    "linear_solver_tolerance": 1e-14,
}

  • 'warm_start: warm starts MDA,

  • 'use_lu_fact: optimize the adjoint resolution by storing the Jacobian matrix LU factorization for the multiple RHS (objective + constraints). This saves CPU time if you can pay for the memory and have the full Jacobians available, not just matrix vector products.

  • 'linear_solver_tolerance': set the linear solver tolerance, idem we need full convergence

design_space = SobieskiDesignSpace()
design_space
Sobieski design space:
Name Lower bound Value Upper bound Type
x_shared[0] 0.01 0.05 0.09 float
x_shared[1] 30000 45000 60000 float
x_shared[2] 1.4 1.6 1.8 float
x_shared[3] 2.5 5.5 8.5 float
x_shared[4] 40 55 70 float
x_shared[5] 500 1000 1500 float
x_1[0] 0.1 0.25 0.4 float
x_1[1] 0.75 1 1.25 float
x_2 0.75 1 1.25 float
x_3 0.1 0.5 1 float
y_14[0] 24850 50606.9741711 77100 float
y_14[1] -7700 7306.20262124 45000 float
y_32 0.235 0.5027962499999999 0.795 float
y_31 2960 6354.32430691 10185 float
y_24 0.44 4.15006276 11.13 float
y_34 0.44 1.10754577 1.98 float
y_23 3365 12194.2671934 26400 float
y_21 24850 50606.9741711 77250 float
y_12[0] 24850 50606.9742 77250 float
y_12[1] 0.45 0.95 1.5 float 


scenario = create_scenario(
    disciplines,
    "y_4",
    design_space,
    maximize_objective=True,
    formulation_name="MDF",
    main_mda_settings=main_mda_settings,
)

WARNING - 08:37:03: Unsupported feature 'minItems' in JSONGrammar 'SobieskiMission_discipline_output' for property 'y_4' in conversion to SimpleGrammar.
WARNING - 08:37:03: Unsupported feature 'maxItems' in JSONGrammar 'SobieskiMission_discipline_output' for property 'y_4' in conversion to SimpleGrammar.

Set the design constraints#

for c_name in ["g_1", "g_2", "g_3"]:
    scenario.add_constraint(c_name, constraint_type="ineq")

XDSMIZE the scenario#

Generate the XDSM file on the fly:

  • log_workflow_status=True will log the status of the workflow in the console,

  • save_html (default True) will generate a self-contained HTML file, that can be automatically opened using show_html=True.

scenario.xdsmize(save_html=False, pdf_build=False)


Define the algorithm inputs#

We set the maximum number of iterations, the optimizer and the optimizer options. Algorithm specific options are passed there. Use the high-level function get_algorithm_options_schema() for more information or read the documentation.

Here the ftol_rel setting is a stop criteria based on the relative difference in the objective between two iterates ineq_tolerance the tolerance determination of the optimum; this is specific to the GEMSEO wrapping and not in the solver.

from gemseo.settings.opt import SLSQP_Settings  # noqa: E402

slsqp_settings = SLSQP_Settings(
    max_iter=10,
    ftol_rel=1e-10,
    ineq_tolerance=2e-3,
    normalize_design_space=True,
)

See also

We can also generate a backup file for the optimization, as well as plots on the fly of the optimization history if option plot is True. This slows down a lot the process, here since SSBJ is very light

scenario.set_optimization_history_backup(
    file_path="mdf_backup.h5",
    at_each_iteration=True,
    at_each_function_call=False,
    erase=True,
    load=False,
    plot=True
)

Execute the scenario#

scenario.execute(slsqp_settings)

   INFO - 08:37:03:
   INFO - 08:37:03: *** Start MDOScenario execution ***
   INFO - 08:37:03: MDOScenario
   INFO - 08:37:03:    Disciplines: SobieskiAerodynamics SobieskiMission SobieskiPropulsion SobieskiStructure
   INFO - 08:37:03:    MDO formulation: MDF
   INFO - 08:37:03: Optimization problem:
   INFO - 08:37:03:    minimize -y_4(x_shared, x_1, x_2, x_3)
   INFO - 08:37:03:    with respect to x_1, x_2, x_3, x_shared
   INFO - 08:37:03:    subject to constraints:
   INFO - 08:37:03:       g_1(x_shared, x_1, x_2, x_3) <= 0
   INFO - 08:37:03:       g_2(x_shared, x_1, x_2, x_3) <= 0
   INFO - 08:37:03:       g_3(x_shared, x_1, x_2, x_3) <= 0
   INFO - 08:37:03:    over the design space:
   INFO - 08:37:03:       +-------------+-------------+-------+-------------+-------+
   INFO - 08:37:03:       | Name        | Lower bound | Value | Upper bound | Type  |
   INFO - 08:37:03:       +-------------+-------------+-------+-------------+-------+
   INFO - 08:37:03:       | x_shared[0] |     0.01    |  0.05 |     0.09    | float |
   INFO - 08:37:03:       | x_shared[1] |    30000    | 45000 |    60000    | float |
   INFO - 08:37:03:       | x_shared[2] |     1.4     |  1.6  |     1.8     | float |
   INFO - 08:37:03:       | x_shared[3] |     2.5     |  5.5  |     8.5     | float |
   INFO - 08:37:03:       | x_shared[4] |      40     |   55  |      70     | float |
   INFO - 08:37:03:       | x_shared[5] |     500     |  1000 |     1500    | float |
   INFO - 08:37:03:       | x_1[0]      |     0.1     |  0.25 |     0.4     | float |
   INFO - 08:37:03:       | x_1[1]      |     0.75    |   1   |     1.25    | float |
   INFO - 08:37:03:       | x_2         |     0.75    |   1   |     1.25    | float |
   INFO - 08:37:03:       | x_3         |     0.1     |  0.5  |      1      | float |
   INFO - 08:37:03:       +-------------+-------------+-------+-------------+-------+
   INFO - 08:37:03: Solving optimization problem with algorithm SLSQP:
   INFO - 08:37:03:     10%|█         | 1/10 [00:00<00:00, 14.49 it/sec, obj=-536]
   INFO - 08:37:03:     20%|██        | 2/10 [00:00<00:00, 11.46 it/sec, obj=-2.12e+3]
   INFO - 08:37:03:     30%|███       | 3/10 [00:00<00:00,  8.75 it/sec, obj=-3.72e+3]
   INFO - 08:37:04:     40%|████      | 4/10 [00:00<00:00,  9.19 it/sec, obj=-3.96e+3]
WARNING - 08:37:04: MDAJacobi has reached its maximum number of iterations but the normed residual 5.126876767779417e-11 is still above the tolerance 1e-14.
   INFO - 08:37:04:     50%|█████     | 5/10 [00:00<00:00,  8.28 it/sec, obj=-3.96e+3]
   INFO - 08:37:04: Optimization result:
   INFO - 08:37:04:    Optimizer info:
   INFO - 08:37:04:       Status: 8
   INFO - 08:37:04:       Message: Positive directional derivative for linesearch
   INFO - 08:37:04:       Number of calls to the objective function by the optimizer: 6
   INFO - 08:37:04:    Solution:
   INFO - 08:37:04:       The solution is feasible.
   INFO - 08:37:04:       Objective: -3963.8622159757956
   INFO - 08:37:04:       Standardized constraints:
   INFO - 08:37:04:          g_1 = [-0.01805379 -0.0333412  -0.04424541 -0.05183129 -0.05732327 -0.13720865
   INFO - 08:37:04:  -0.10279135]
   INFO - 08:37:04:          g_2 = 9.423096252181296e-07
   INFO - 08:37:04:          g_3 = [-0.76777633 -0.23222367  0.00080718 -0.183255  ]
   INFO - 08:37:04:       Design space:
   INFO - 08:37:04:          +-------------+-------------+---------------------+-------------+-------+
   INFO - 08:37:04:          | Name        | Lower bound |        Value        | Upper bound | Type  |
   INFO - 08:37:04:          +-------------+-------------+---------------------+-------------+-------+
   INFO - 08:37:04:          | x_shared[0] |     0.01    | 0.06000023557740629 |     0.09    | float |
   INFO - 08:37:04:          | x_shared[1] |    30000    |        60000        |    60000    | float |
   INFO - 08:37:04:          | x_shared[2] |     1.4     |         1.4         |     1.8     | float |
   INFO - 08:37:04:          | x_shared[3] |     2.5     |         2.5         |     8.5     | float |
   INFO - 08:37:04:          | x_shared[4] |      40     |          70         |      70     | float |
   INFO - 08:37:04:          | x_shared[5] |     500     |         1500        |     1500    | float |
   INFO - 08:37:04:          | x_1[0]      |     0.1     |         0.4         |     0.4     | float |
   INFO - 08:37:04:          | x_1[1]      |     0.75    |         0.75        |     1.25    | float |
   INFO - 08:37:04:          | x_2         |     0.75    |         0.75        |     1.25    | float |
   INFO - 08:37:04:          | x_3         |     0.1     |  0.1563708636523273 |      1      | float |
   INFO - 08:37:04:          +-------------+-------------+---------------------+-------------+-------+
   INFO - 08:37:04: *** End MDOScenario execution (time: 0:00:00.646958) ***

Save the optimization history#

We can save the whole optimization problem and its history for further post-processing:

scenario.save_optimization_history("mdf_history.h5", file_format="hdf5")

INFO - 08:37:04: Exporting the optimization problem to the file mdf_history.h5 at node

We can also save only calls to functions and design variables history:

scenario.save_optimization_history("mdf_history.xml", file_format="ggobi")

Post-process the results#

Plot the optimization history view#

scenario.post_process(post_name="OptHistoryView", save=False, show=True)
  • Evolution of the optimization variables
  • Evolution of the objective value
  • Evolution of the distance to the optimum
  • Evolution of the inequality constraints
<gemseo.post.opt_history_view.OptHistoryView object at 0x7f6dc82d04c0>

Note that post-processor settings passed to BaseScenario.post_process can be provided via a Pydantic model (see the example below). For more information, see Post-processor Settings.

Plot the basic history view#

from gemseo.settings.post import BasicHistory_Settings  # noqa: E402

scenario.post_process(
    BasicHistory_Settings(variable_names=["x_shared"], save=False, show=True)
)
History plot
<gemseo.post.basic_history.BasicHistory object at 0x7f6e058f3bb0>

Plot the constraints and objective history#

scenario.post_process(post_name="ObjConstrHist", save=False, show=True)
Evolution of the objective and maximum constraint
<gemseo.post.obj_constr_hist.ObjConstrHist object at 0x7f6dc1113760>

Plot the constraints history#

scenario.post_process(
    post_name="ConstraintsHistory",
    constraint_names=["g_1", "g_2", "g_3"],
    save=False,
    show=True,
)
Evolution of the constraints w.r.t. iterations, g_1[0] (inequality), g_1[1] (inequality), g_1[2] (inequality), g_1[3] (inequality), g_1[4] (inequality), g_1[5] (inequality), g_1[6] (inequality), g_2 (inequality), g_3[0] (inequality), g_3[1] (inequality), g_3[2] (inequality), g_3[3] (inequality)
<gemseo.post.constraints_history.ConstraintsHistory object at 0x7f6dc1e6d100>

Plot the constraints history using a radar chart#

scenario.post_process(
    post_name="RadarChart",
    constraint_names=["g_1", "g_2", "g_3"],
    save=False,
    show=True,
)
Constraints at iteration 4 (optimum)
/home/docs/checkouts/readthedocs.org/user_builds/gemseo/envs/6.0.0/lib/python3.9/site-packages/gemseo/post/dataset/plots/_matplotlib/plot.py:87: UserWarning: This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.
  figure.tight_layout()

<gemseo.post.radar_chart.RadarChart object at 0x7f6dc11cba30>

Plot the quadratic approximation of the objective#

scenario.post_process(post_name="QuadApprox", function="-y_4", save=False, show=True)
  • Hessian matrix SR1 approximation of -y_4
  • plot sobieski mdf example
<gemseo.post.quad_approx.QuadApprox object at 0x7f6dc11dc490>

Plot the functions using a SOM#

scenario.post_process(post_name="SOM", save=False, show=True)
Self Organizing Maps of the design space, -y_4, g_1[0], g_1[1], g_1[2], g_1[3], g_1[4], g_1[5], g_1[6], g_2, g_3[0], g_3[1], g_3[2], g_3[3]
    INFO - 08:37:08: Building Self Organizing Map from optimization history:
    INFO - 08:37:08:     Number of neurons in x direction = 4
    INFO - 08:37:08:     Number of neurons in y direction = 4

<gemseo.post.som.SOM object at 0x7f6dc1c1c370>

Plot the scatter matrix of variables of interest#

scenario.post_process(
    post_name="ScatterPlotMatrix",
    variable_names=["-y_4", "g_1"],
    save=False,
    show=True,
    fig_size=(14, 14),
)
plot sobieski mdf example
<gemseo.post.scatter_plot_matrix.ScatterPlotMatrix object at 0x7f6dc1cb8850>

Plot the variables using the parallel coordinates#

scenario.post_process(post_name="ParallelCoordinates", save=False, show=True)
  • Design variables history colored by '-y_4' value
  • Objective function and constraints history colored by '-y_4' value.
<gemseo.post.parallel_coordinates.ParallelCoordinates object at 0x7f6df81bee20>

Plot the robustness of the solution#

scenario.post_process(post_name="Robustness", save=False, show=True)
Boxplot of the optimization functions with normalized stddev 0.01
<gemseo.post.robustness.Robustness object at 0x7f6dc19c25e0>

Plot the influence of the design variables#

scenario.post_process(
    post_name="VariableInfluence", fig_size=(14, 14), save=False, show=True
)
9 variables explain 99% of -y_4, 5 variables explain 99% of g_1[0], 5 variables explain 99% of g_1[1], 5 variables explain 99% of g_1[2], 5 variables explain 99% of g_1[3], 5 variables explain 99% of g_1[4], 4 variables explain 99% of g_1[5], 4 variables explain 99% of g_1[6], 1 variables explain 99% of g_2, 7 variables explain 99% of g_3[0], 7 variables explain 99% of g_3[1], 3 variables explain 99% of g_3[2], 3 variables explain 99% of g_3[3]
    INFO - 08:37:13: Output name; most influential variables to explain 0.99% of the output variation
    INFO - 08:37:13:    -y_4; x_1[1], x_2, x_3, x_shared[0], x_shared[1], x_shared[2], x_shared[3], x_shared[4], x_shared[5]
    INFO - 08:37:13:    g_1[0]; x_1[0], x_1[1], x_shared[0], x_shared[3], x_shared[5]
    INFO - 08:37:13:    g_1[1]; x_1[0], x_1[1], x_shared[0], x_shared[3], x_shared[5]
    INFO - 08:37:13:    g_1[2]; x_1[0], x_1[1], x_shared[0], x_shared[3], x_shared[5]
    INFO - 08:37:13:    g_1[3]; x_1[0], x_1[1], x_shared[0], x_shared[3], x_shared[5]
    INFO - 08:37:13:    g_1[4]; x_1[0], x_1[1], x_shared[0], x_shared[3], x_shared[5]
    INFO - 08:37:13:    g_1[5]; x_1[0], x_1[1], x_shared[3], x_shared[5]
    INFO - 08:37:14:    g_1[6]; x_1[0], x_1[1], x_shared[3], x_shared[5]
    INFO - 08:37:14:    g_2; x_shared[0]
    INFO - 08:37:14:    g_3[0]; x_2, x_3, x_shared[0], x_shared[1], x_shared[2], x_shared[4], x_shared[5]
    INFO - 08:37:14:    g_3[1]; x_2, x_3, x_shared[0], x_shared[1], x_shared[2], x_shared[4], x_shared[5]
    INFO - 08:37:14:    g_3[2]; x_3, x_shared[1], x_shared[2]
    INFO - 08:37:14:    g_3[3]; x_3, x_shared[1], x_shared[2]

<gemseo.post.variable_influence.VariableInfluence object at 0x7f6dc04d4610>

Total running time of the script: (0 minutes 12.107 seconds)

Gallery generated by Sphinx-Gallery