Comparing sensitivity indices

from __future__ import annotations

from gemseo import configure_logger
from gemseo.problems.uncertainty.ishigami.ishigami_discipline import IshigamiDiscipline
from gemseo.problems.uncertainty.ishigami.ishigami_space import IshigamiSpace
from gemseo.uncertainty.sensitivity.correlation_analysis import CorrelationAnalysis
from gemseo.uncertainty.sensitivity.morris_analysis import MorrisAnalysis

configure_logger()

<RootLogger root (INFO)>

In this example, we consider the Ishigami function [IH90]

\[f(x_1,x_2,x_3)=\sin(x_1)+7\sin(x_2)^2+0.1x_3^4\sin(x_1)\]

implemented as an Discipline by the IshigamiDiscipline. It is commonly used with the independent random variables \(X_1\), \(X_2\) and \(X_3\) uniformly distributed between \(-\pi\) and \(\pi\) and defined in the IshigamiSpace.

discipline = IshigamiDiscipline()
uncertain_space = IshigamiSpace()

We would like to carry out two sensitivity analyses, e.g. a first one based on correlation coefficients and a second one based on the Morris methodology, and compare the results,

Firstly, we create a CorrelationAnalysis and compute the sensitivity indices:

correlation = CorrelationAnalysis()
correlation.compute_samples([discipline], uncertain_space, 10)
correlation.compute_indices()

 WARNING - 08:35:10: No coupling in MDA, switching chain_linearize to True.
    INFO - 08:35:10:
    INFO - 08:35:10: *** Start CorrelationAnalysisSamplingPhase execution ***
    INFO - 08:35:10: CorrelationAnalysisSamplingPhase
    INFO - 08:35:10:    Disciplines: IshigamiDiscipline
    INFO - 08:35:10:    MDO formulation: MDF
    INFO - 08:35:10: Running the algorithm OT_MONTE_CARLO:
    INFO - 08:35:10:     10%|█         | 1/10 [00:00<00:00, 356.75 it/sec]
    INFO - 08:35:10:     20%|██        | 2/10 [00:00<00:00, 579.72 it/sec]
    INFO - 08:35:10:     30%|███       | 3/10 [00:00<00:00, 737.14 it/sec]
    INFO - 08:35:10:     40%|████      | 4/10 [00:00<00:00, 858.83 it/sec]
    INFO - 08:35:10:     50%|█████     | 5/10 [00:00<00:00, 956.64 it/sec]
    INFO - 08:35:10:     60%|██████    | 6/10 [00:00<00:00, 1031.60 it/sec]
    INFO - 08:35:10:     70%|███████   | 7/10 [00:00<00:00, 1095.94 it/sec]
    INFO - 08:35:10:     80%|████████  | 8/10 [00:00<00:00, 1149.71 it/sec]
    INFO - 08:35:10:     90%|█████████ | 9/10 [00:00<00:00, 1195.75 it/sec]
    INFO - 08:35:10:    100%|██████████| 10/10 [00:00<00:00, 1227.12 it/sec]
    INFO - 08:35:10: *** End CorrelationAnalysisSamplingPhase execution (time: 0:00:00.013713) ***

CorrelationAnalysis.SensitivityIndices(kendall={'y': [{'x1': array([0.55555556]), 'x2': array([0.02222222]), 'x3': array([-0.11111111])}]}, pcc={'y': [{'x1': array([0.84696461]), 'x2': array([0.68814608]), 'x3': array([-0.29846394])}]}, pearson={'y': [{'x1': array([0.685388]), 'x2': array([0.09681897]), 'x3': array([-0.23027298])}]}, prcc={'y': [{'x1': array([0.90374102]), 'x2': array([0.76539572]), 'x3': array([-0.02232206])}]}, spearman={'y': [{'x1': array([0.74545455]), 'x2': array([0.04242424]), 'x3': array([-0.09090909])}]}, src={'y': [{'x1': array([0.94001308]), 'x2': array([0.55748872]), 'x3': array([-0.16157012])}]}, srrc={'y': [{'x1': array([1.06252802]), 'x2': array([0.60167726]), 'x3': array([-0.00959941])}]}, ssrc={'y': [{'x1': array([0.88362459]), 'x2': array([0.31079367]), 'x3': array([0.0261049])}]})

Then, we create an MorrisAnalysis and compute the sensitivity indices:

morris = MorrisAnalysis()
morris.compute_samples([discipline], uncertain_space, 10)
morris.compute_indices()

 WARNING - 08:35:10: No coupling in MDA, switching chain_linearize to True.
    INFO - 08:35:10:
    INFO - 08:35:10: *** Start MorrisAnalysisSamplingPhase execution ***
    INFO - 08:35:10: MorrisAnalysisSamplingPhase
    INFO - 08:35:10:    Disciplines: IshigamiDiscipline
    INFO - 08:35:10:    MDO formulation: MDF
    INFO - 08:35:10: Running the algorithm MorrisDOE:
    INFO - 08:35:10:     12%|█▎        | 1/8 [00:00<00:00, 1940.91 it/sec]
    INFO - 08:35:10:     25%|██▌       | 2/8 [00:00<00:00, 1741.82 it/sec]
    INFO - 08:35:10:     38%|███▊      | 3/8 [00:00<00:00, 1705.23 it/sec]
    INFO - 08:35:10:     50%|█████     | 4/8 [00:00<00:00, 1681.76 it/sec]
    INFO - 08:35:10:     62%|██████▎   | 5/8 [00:00<00:00, 1678.93 it/sec]
    INFO - 08:35:10:     75%|███████▌  | 6/8 [00:00<00:00, 1689.44 it/sec]
    INFO - 08:35:10:     88%|████████▊ | 7/8 [00:00<00:00, 1688.53 it/sec]
    INFO - 08:35:10:    100%|██████████| 8/8 [00:00<00:00, 1681.17 it/sec]
    INFO - 08:35:10: *** End MorrisAnalysisSamplingPhase execution (time: 0:00:00.010322) ***

MorrisAnalysis.SensitivityIndices(mu={'y': [{'x1': array([0.73532408]), 'x2': array([-0.05115399]), 'x3': array([-1.6024484])}]}, mu_star={'y': [{'x1': array([0.76770333]), 'x2': array([2.09435091]), 'x3': array([1.6024484])}]}, sigma={'y': [{'x1': array([0.76770333]), 'x2': array([2.09435091]), 'x3': array([1.58984353])}]}, relative_sigma={'y': [{'x1': array([1.]), 'x2': array([1.]), 'x3': array([0.99213399])}]}, min={'y': [{'x1': array([0.03237925]), 'x2': array([2.04319692]), 'x3': array([0.01260487])}]}, max={'y': [{'x1': array([1.50302741]), 'x2': array([2.14550491]), 'x3': array([3.19229192])}]})

Lastly, we compare these analyses with the graphical method BaseSensitivityAnalysis.plot_comparison(), either using a bar chart:

morris.plot_comparison(correlation, "y", use_bar_plot=True, save=False, show=True)

<gemseo.post.dataset.bars.BarPlot object at 0x7f6e05adc2b0>

or a radar plot:

morris.plot_comparison(correlation, "y", use_bar_plot=False, save=False, show=True)

/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.dataset.radar_chart.RadarChart object at 0x7f6e05ccb070>