SobolIndicesAlgorithm

(Source code, png, hires.png, pdf)

../../_images/SobolIndicesAlgorithm.png
class SobolIndicesAlgorithm(*args)

Sensitivity analysis.

Notes

This method analyzes the influence of each component of an input random vector \vect{X} = \left( X^1, \ldots, X^{n_X} \right) on an output random vector \vect{Y} = \left( Y^1, \ldots, Y^{n_Y} \right) by computing Sobol’ indices (see [sobol1993]). It computes, for every output random variable Y^k (1 \leqslant k \leqslant n_Y), the part of its variance due to each input component X^i (1 \leqslant i \leqslant n_X) of \vect{X}.

Let \vect{g} be the physical model such as \vect{Y}=\vect{g}(\vect{X}). Let us first consider the case where n_Y=1. In that case, we denote \vect{Y} by Y. Let us write the variance of Y as a function of \vect{X} = \left( X^1, \ldots, X^{n_X} \right). For any subset I \subseteq \{1, \ldots, n_X\}, define \vect{X}_I = \left( X^i \right)_{i \in I}.

Using the Hoeffding decomposition, we get:

\Var{Y} = \sum_{ I \subseteq \{1, \ldots, n_X\} } V_I

where:

V_I = \Var{ \sum_{J \subseteq I} (-1)^{ |I| - |J| } \Expect{ Y \vert \mat{X}_J} }.

For the sake of conciseness, for any integers 1 \leqslant i, j \leqslant n_X, define V_i = V_{\{i\}} and V_{i,j} = V_{\{i,j\}}. We have:

V_i     & = \Var{ \Expect{Y \vert X_i} } \\
V_{i,j} & = \Var{ \Expect{ Y \vert X_i, X_j}  - \Expect{Y \vert X_i} - \Expect{Y \vert X_j} } \\
        & = \Var{ \Expect{ Y \vert X_i, X_j} } - V_i - V_j

for i=1,\ldots,n_X.

First and second order Sobol’ indices are defined as follows:

S_i     & = \frac{V_i}{\Var{Y}} \\
S_{i,j} & = \frac{V_{i,j}}{\Var{Y}}

for i=1,\ldots,n_X.

The first order Sobol’ index S_i measures the part of the variance of Y explained by X_i alone. The second order Sobol’ index S_{i,j} measures the part of the variance of Y explained by the interaction of X_i and X_j.

For any i=1,\ldots,n_X, let us define

VT_i   & = \sum_{I \ni i} V_I \\
V_{-i} & = \Var{ \Expect{Y \vert X_1, \ldots, X_{i-1}, X_{i+1}, \ldots X_{n_X}} }.

Total order Sobol’ indices are defined as follows:

ST_i = \frac{VT_i}{\Var{Y}} = 1 - \frac{V_{-i}}{\Var{Y}}

for i=1,\ldots,n_X.

The total order Sobol’ index ST_i quantifies the part of the variance of Y that is due to the effect of X_i and its interactions with all the other input variables. It can also be viewed as the part of the variance of Y that cannot be explained without X_i.

In practice, to estimate these quantities, Sobol’ proposes to use numerical methods that rely on two independent realizations of the random vector \vect{X}. Let \mat{A} and \mat{B} be two independent samples (of size N) of \vect{X} :

\mat{A} = \left(
\begin{array}{cccc}
a_{1,1} & a_{1,2} & \cdots & a_{1, n_X} \\
a_{2,1} & a_{2,2} & \cdots & a_{2, n_X} \\
\vdots  & \vdots  & \ddots  & \vdots \\
a_{N,1} & a_{1,2} & \cdots & a_{N, n_X}
\end{array}
\right), \  \mat{B} = \left(
\begin{array}{cccc}
b_{1,1} & b_{1,2} & \cdots & b_{1, n_X} \\
b_{2,1} & b_{2,2} & \cdots & b_{2, n_X} \\
\vdots  & \vdots  & \vdots  & \vdots \\
b_{N,1} & b_{1,2} & \cdots & b_{N, n_X}
\end{array}
\right)

Each line is a realization of the random vector \vect{X}.

We are now going to mix these two samples to get an estimate of the sensitivity indices.

\mat{E}^i = \left(
\begin{array}{cccccc}
a_{1,1} & a_{1,2} & \cdots & b_{1,i} & \cdots & a_{1, n_X} \\
a_{2,1} & a_{2,2} & \cdots & b_{2,i} & \cdots & a_{2, n_X} \\
\vdots  & \vdots  &        & \vdots  & \ddots & \vdots \\
a_{N,1} & a_{1,2} & \cdots & b_{N,i} & \cdots & a_{N, n_X}
\end{array}
\right),
\mat{C}^i = \left(
\begin{array}{cccccc}
b_{1,1} & b_{1,2} & \cdots & a_{1,i} & \cdots & b_{1, n_X} \\
b_{2,1} & b_{2,2} & \cdots & a_{2,i} & \cdots & b_{2, n_X} \\
\vdots  & \vdots  &        & \vdots  & \ddots  & \vdots \\
b_{N,1} & b_{1,2} & \cdots & a_{N,i} & \cdots & b_{N, n_X}
\end{array}
\right)

For the sake of stability, computations will be performed with centered output. Let \overline{\vect{g}} be the mean of the combined samples \vect{g}(\mat{A}) and \vect{g}(\mat{B}). Let \tilde{\vect{g}} be the empirically centered function defined, for any \vect{x} \in \Rset^{n_X}, by:

\tilde{\vect{g}}(\vect{x}) = \vect{g}(\vect{x}) - \overline{\vect{g}}.

To estimate the total variance \Var{Y}, we use the computeVariance() method of the Sample \tilde{g}(\mat{A}).

Several estimators of V_i, VT_i and V_{-i} are provided by the SobolIndicesAlgorithm implementations:

Specific formulas for \hat{V}_i, \widehat{VT}_i and \hat{V}_{-i} are given in the corresponding documentation pages.

The estimator \hat{V}_{i,j} of V_{i,j} is the same for all these classes:

\hat{V}_{i,j} = \frac{1}{N-1} \sum_{k=1}^{N} \tilde{g}(\vect{E}_k^i) \tilde{g}(\vect{C}_k^j) - \frac{1}{N} \sum_{k=1}^{N} \tilde{g}(\vect{A}_k) \tilde{g}(\vect{B}_k) - \hat{V}_i - \hat{V}_j.

Notice that the value of the second order conditional variance depends on the estimators \hat{V}_i and \hat{V}_j which are method-dependent. This implies that the value of the second order indices may depend on the specific Sobol’ estimator we use.

For multivariate outputs i.e. when n_Y>1, we compute the Sobol’ indices with respect to each output variable. In this case, the methods getFirstOrderIndices() and getTotalOrderIndices() return the Sobol’ indices of the first output, but the index of the output can be specified as input argument. Moreover, the indices can be aggregated [gamboa2013]. Let V_i^{(k)} be the (first order) variance of the conditional expectation of the k-th output Y^{(k)}:

V_i^{(k)} & = \Var{ \Expect{Y^{(k)} \vert X_i} }

for i=1,\ldots,n_X and k=1,\ldots,n_Y. Similarily, let VT_i^{(k)} be the total variance of the conditional expectation of Y^{(k)} for i=1,\ldots,n_X and k=1,\ldots,n_Y.

The indices can be aggregated with the following formulas:

S_i^{(a)}  & =  \frac{ \sum_{k=1}^{n_Y} V_{i}^{(k)} }{ \sum_{k=1}^{n_Y} \Var{Y_k} }  \\
ST_i^{(a)} & =  \frac{ \sum_{k=1}^{n_Y} VT_{i}^{(k)} }{ \sum_{k=1}^{n_Y} \Var{Y_k} }

for i=1,\ldots,n_X.

Aggregated indices can be retrieved with the getAggregatedFirstOrderIndices() and getAggregatedTotalOrderIndices() methods.


Notice that the distribution of the estimators of the first and total order indices can be estimated thanks to the getFirstOrderIndicesDistribution() and getTotalOrderIndicesDistribution() methods. This is done either through bootstrapping or using an asymptotic estimator. The ResourceMap key SobolIndicesAlgorithm-DefaultUseAsymptoticDistribution stores a boolean that decides the default behavior, but it can be overridden by the method setUseAsymptoticDistribution().

Corresponding confidence intervals are provided by the methods getFirstOrderIndicesInterval() and getTotalOrderIndicesInterval(). Their confidence level can be adjusted with setConfidenceLevel(). The default confidence level is stored in the ResourceMap and can be accessed with the SobolIndicesAlgorithm-DefaultConfidenceLevel key.

The asymptotic estimator of the distribution requires an asymptotic estimate of its variance, which is computed using the [janon2014] delta method, as expained in the technical report [pmfre01116].

Methods

DrawCorrelationCoefficients(*args)

Draw the correlation coefficients.

DrawImportanceFactors(*args)

Draw the importance factors.

DrawSobolIndices(*args)

Draw the Sobol’ indices.

draw(*args)

Draw sensitivity indices.

getAggregatedFirstOrderIndices()

Get the evaluation of aggregated first order Sobol indices.

getAggregatedTotalOrderIndices()

Get the evaluation of aggregated total order Sobol indices.

getBootstrapSize()

Get the number of bootstrap sampling size.

getClassName()

Accessor to the object’s name.

getConfidenceLevel()

Get the confidence interval level for confidence intervals.

getFirstOrderIndices([marginalIndex])

Get first order Sobol indices.

getFirstOrderIndicesDistribution()

Get the distribution of the aggregated first order Sobol indices.

getFirstOrderIndicesInterval()

Get interval for the aggregated first order Sobol indices.

getId()

Accessor to the object’s id.

getImplementation()

Accessor to the underlying implementation.

getName()

Accessor to the object’s name.

getSecondOrderIndices([marginalIndex])

Get second order Sobol indices.

getTotalOrderIndices([marginalIndex])

Get total order Sobol indices.

getTotalOrderIndicesDistribution()

Get the distribution of the aggregated total order Sobol indices.

getTotalOrderIndicesInterval()

Get interval for the aggregated total order Sobol indices.

getUseAsymptoticDistribution()

Select asymptotic or bootstrap confidence intervals.

setBootstrapSize(bootstrapSize)

Set the number of bootstrap sampling size.

setConfidenceLevel(confidenceLevel)

Set the confidence interval level for confidence intervals.

setDesign(inputDesign, outputDesign, size)

Sample accessor.

setName(name)

Accessor to the object’s name.

setUseAsymptoticDistribution(…)

Select asymptotic or bootstrap confidence intervals.

__init__(*args)

Initialize self. See help(type(self)) for accurate signature.

static DrawCorrelationCoefficients(*args)
Draw the correlation coefficients.

As correlation coefficients are considered, values might be positive or negative.

Available usages:

DrawCorrelationCoefficients(correlationCoefficients, title=’Correlation coefficients’)

DrawCorrelationCoefficients(values, names, title=’Correlation coefficients’)

Parameters
correlationCoefficientsPointWithDescription

Sequence containing the correlation coefficients with a description for each component. The descriptions are used to build labels for the created graph. If they are not mentioned, default labels will be used.

valuessequence of float

Correlation coefficients.

namessequence of str

Variables’ names used to build labels for the created the graph.

titlestr

Title of the graph.

Returns
graphGraph

A graph containing a Cloud and a Text of the correlation coefficients.

static DrawImportanceFactors(*args)

Draw the importance factors.

Available usages:

DrawImportanceFactors(importanceFactors, title=’Importance Factors’)

DrawImportanceFactors(values, names, title=’Importance Factors’)

Parameters
importanceFactorsPointWithDescription

Sequence containing the importance factors with a description for each component. The descriptions are used to build labels for the created Pie. If they are not mentioned, default labels will be used.

valuessequence of float

Importance factors.

namessequence of str

Variables’ names used to build labels for the created Pie.

titlestr

Title of the graph.

Returns
graphGraph

A graph containing a Pie of the importance factors of the variables.

static DrawSobolIndices(*args)

Draw the Sobol’ indices.

Parameters
inputDescriptionsequence of str

Variable names

firstOrderIndicessequence of float

First order indices values

totalOrderIndicessequence of float

Total order indices values

fo_ciInterval, optional

First order indices confidence interval

to_ciInterval, optional

Total order indices confidence interval

Returns
graphGraph

For each variable, draws first and total indices

draw(*args)

Draw sensitivity indices.

Usage:

draw()

draw(marginalIndex)

With the first usage, draw the aggregated first and total order indices. With the second usage, draw the first and total order indices of a specific marginal in case of vectorial output

Parameters
marginalIndex: int

marginal of interest (case of second usage)

Returns
graphGraph

A graph containing the aggregated first and total order indices.

Notes

If number of bootstrap sampling is not 0, and confidence level associated > 0, the graph includes confidence interval plots in the first usage.

getAggregatedFirstOrderIndices()

Get the evaluation of aggregated first order Sobol indices.

Returns
indicesPoint

Sequence containing aggregated first order Sobol indices.

getAggregatedTotalOrderIndices()

Get the evaluation of aggregated total order Sobol indices.

Returns
indicesPoint

Sequence containing aggregated total order Sobol indices.

getBootstrapSize()

Get the number of bootstrap sampling size.

Returns
bootstrapSizeint

Number of bootsrap sampling

getClassName()

Accessor to the object’s name.

Returns
class_namestr

The object class name (object.__class__.__name__).

getConfidenceLevel()

Get the confidence interval level for confidence intervals.

Returns
confidenceLevelfloat

Confidence level for confidence intervals

getFirstOrderIndices(marginalIndex=0)

Get first order Sobol indices.

Parameters
iint, optional

Index of the output marginal of the function, equal to 0 by default.

Returns
indicesPoint

Sequence containing first order Sobol indices.

getFirstOrderIndicesDistribution()

Get the distribution of the aggregated first order Sobol indices.

Returns
distributionDistribution

Distribution for first order Sobol indices for each component.

getFirstOrderIndicesInterval()

Get interval for the aggregated first order Sobol indices.

Returns
intervalInterval

Interval for first order Sobol indices for each component. Computed marginal by marginal (not from the joint distribution).

getId()

Accessor to the object’s id.

Returns
idint

Internal unique identifier.

getImplementation()

Accessor to the underlying implementation.

Returns
implImplementation

The implementation class.

getName()

Accessor to the object’s name.

Returns
namestr

The name of the object.

getSecondOrderIndices(marginalIndex=0)

Get second order Sobol indices.

Parameters
iint, optional

Index of the marginal of the function, equals to 0 by default.

Returns
indicesSymmetricMatrix

Tensor containing second order Sobol indices.

getTotalOrderIndices(marginalIndex=0)

Get total order Sobol indices.

Parameters
iint, optional

Index of the output marginal of the function, equal to 0 by default.

Returns
indicesPoint

Sequence containing total order Sobol indices.

getTotalOrderIndicesDistribution()

Get the distribution of the aggregated total order Sobol indices.

Returns
distributionDistribution

Distribution for total order Sobol indices for each component.

getTotalOrderIndicesInterval()

Get interval for the aggregated total order Sobol indices.

Returns
intervalInterval

Interval for total order Sobol indices for each component. Computed marginal by marginal (not from the joint distribution).

getUseAsymptoticDistribution()

Select asymptotic or bootstrap confidence intervals.

Returns
useAsymptoticDistributionbool

Whether to use bootstrap or asymptotic intervals

setBootstrapSize(bootstrapSize)

Set the number of bootstrap sampling size.

Default value is 0.

Parameters
bootstrapSizeint

Number of bootsrap sampling

setConfidenceLevel(confidenceLevel)

Set the confidence interval level for confidence intervals.

Parameters
confidenceLevelfloat

Confidence level for confidence intervals

setDesign(inputDesign, outputDesign, size)

Sample accessor.

Parameters
inputDesignSample

Design for the evaluation of sensitivity indices, obtained thanks to the SobolIndicesAlgorithmImplementation.Generate method

outputDesignSample

Design for the evaluation of sensitivity indices, obtained as the evaluation of a Function (model) on the previous inputDesign

Nint

Size of samples to generate

setName(name)

Accessor to the object’s name.

Parameters
namestr

The name of the object.

setUseAsymptoticDistribution(useAsymptoticDistribution)

Select asymptotic or bootstrap confidence intervals.

Default value is set by the SobolIndicesAlgorithm-DefaultUseAsymptoticDistribution key.

Parameters
useAsymptoticDistributionbool

Whether to use bootstrap or asymptotic intervals