Global Sensitivity Analysis

Global Sensitivity Analysis (GSA) methods are used to quantify the uncertainty in output of a model w.r.t. the parameters, their individual contributions, or the contribution of their interactions. The GSA interface allows for utilizing batched functions for parallel computation of GSA quantities.


This functionality does not come standard with DifferentialEquations.jl. To use this functionality, you must install DiffEqSensitivity.jl:

]add DiffEqSensitivity
using DiffEqSensitivity

General Interface

The general interface for calling a global sensitivity analysis is either:

effects = gsa(f, method, param_range; N, batch=false)


  • y=f(x) is a function that takes in a single vector and spits out a single vector or scalar. If batch=true, then f takes in a matrix where each row is a set of parameters, and returns a matrix where each row is a the output for the corresponding row of parameters.
  • method is one of the GSA methods below.
  • param_range is a vector of tuples for the upper and lower bound for the given parameter i.
  • N is a required keyword argument for the number of samples to take in the trajectories/design.

Note that for some methods there is a second interface where one can directly pass the design matrices:

effects = gsa(f, method, A, B; batch=false)

where A and B are design matrices with each row being a set of parameters. Note that generate_design_matrices from QuasiMonteCarlo.jl can be used to generate the design matrices.

Morris Method

Morris has the following keyword arguments:

  • p_steps - Vector of $\Delta$ for the step sizes in each direction. Required.
  • relative_scale - The elementary effects are calculated with the assumption that the parameters lie in the range [0,1] but as this is not always the case scaling is used to get more informative, scaled effects. Defaults to false.
  • total_num_trajectory, num_trajectory - The total number of design matrices that are generated out of which num_trajectory matrices with the highest spread are used in calculation.
  • len_design_mat - The size of a design matrix.

Morris Method Details

The Morris method also known as Morris’s OAT method where OAT stands for One At a Time can be described in the following steps:

We calculate local sensitivity measures known as “elementary effects”, which are calculated by measuring the perturbation in the output of the model on changing one parameter.

$EE_i = \frac{f(x_1,x_2,..x_i+ \Delta,..x_k) - y}{\Delta}$

These are evaluated at various points in the input chosen such that a wide “spread” of the parameter space is explored and considered in the analysis, to provide an approximate global importance measure. The mean and variance of these elementary effects is computed. A high value of the mean implies that a parameter is important, a high variance implies that its effects are non-linear or the result of interactions with other inputs. This method does not evaluate separately the contribution from the interaction and the contribution of the parameters individually and gives the effects for each parameter which takes into consideration all the interactions and its individual contribution.

Sobol Method

The Sobol object has as its fields the order of the indices to be estimated.

  • order - the order of the indices to calculate. Defaults to [0,1], which means the Total and First order indices. Passing 2 enables calculation of the Second order indices as well.
  • Ei_estimator - Can take :Homma1996, :Sobol2007 and :Jansen1999 for which Monte Carlo estimator is used for the Ei term. Defaults to :Jansen1999.

Sobol Method Details

Sobol is a variance-based method and it decomposes the variance of the output of the model or system into fractions which can be attributed to inputs or sets of inputs. This helps to get not just the individual parameter's sensitivities but also gives a way to quantify the affect and sensitivity from the interaction between the parameters.

\[ Y = f_0+ \sum_{i=1}^d f_i(X_i)+ \sum_{i < j}^d f_{ij}(X_i,X_j) ... + f_{1,2...d}(X_1,X_2,..X_d)\]

\[ Var(Y) = \sum_{i=1}^d V_i + \sum_{i < j}^d V_{ij} + ... + V_{1,2...,d}\]

The Sobol Indices are "order"ed, the first order indices given by $S_i = \frac{V_i}{Var(Y)}$ the contribution to the output variance of the main effect of $X_i$, therefore it measures the effect of varying $X_i$ alone, but averaged over variations in other input parameters. It is standardised by the total variance to provide a fractional contribution. Higher-order interaction indices $S_{i,j}, S_{i,j,k}$ and so on can be formed by dividing other terms in the variance decomposition by $Var(Y)$.

eFAST Method

eFAST has num_harmonics as the only argument, it is the number of harmonics to sum in the Fourier series decomposition and defaults to 4.

eFAST Method Details

eFAST offers a robust, especially at low sample size, and computationally efficient procedure to get the first and total order indices as discussed in Sobol. It utilizes monodimensional Fourier decomposition along a curve exploring the parameter space. The curve is defined by a set of parametric equations,

\[x_{i}(s) = G_{i}(\sin ω_{i}s), ∀ i=1,2 ,..., n,\]

where s is a scalar variable varying over the range $-∞ < s < +∞$, $G_{i}$ are transformation functions and ${ω_{i}}, ∀ i=1,2,...,n$ is a set of different (angular) frequencies, to be properly selected, associated with each factor. For more details on the transformation used and other implementation details you can go through A. Saltelli et al..

Regression Method

RegressionGSA has the following keyword arguments:

  • rank: flag which determines whether to calculate the rank coefficients. Defaults to false.

It returns a RegressionGSAResult, which contains the pearson, standard_regression, and partial_correlation coefficients, described below. If rank is true, then it also contains the ranked versions of these coefficients. Note that the ranked version of the pearson coefficient is also known as the Spearman coefficient, which is returned here as the pearson_rank coefficient.

For multi-variable models, the coefficient for the $X_i$ input variable relating to the $Y_j$ output variable is given as the [i, j] entry in the corresponding returned matrix.

Regression Details

It is possible to fit a linear model explaining the behavior of Y given the values of X, provided that the sample size n is sufficiently large (at least n > d).

The measures provided for this analysis by us in DiffEqSensitivity.jl are

a) Pearson Correlation Coefficient:

\[r = \frac{\sum_{i=1}^{n} (x_i - \overline{x})(y_i - \overline{y})}{\sqrt{\sum_{i=1}^{n} (x_i - \overline{x})^2(y_i - \overline{y})^2}}\]

b) Standard Regression Coefficient (SRC):

\[SRC_j = \beta_{j} \sqrt{\frac{Var(X_j)}{Var(Y)}}\]

where $\beta_j$ is the linear regression coefficient associated to $X_j$. This is also known as a sigma-normalized derivative.

c) Partial Correlation Coefficient (PCC):

\[PCC_j = \rho(X_j - \hat{X_{-j}},Y_j - \hat{Y_{-j}})\]

where $\hat{X_{-j}}$ is the prediction of the linear model, expressing $X_{j}$ with respect to the other inputs and $\hat{Y_{-j}}$ is the prediction of the linear model where $X_j$ is absent. PCC measures the sensitivity of $Y$ to $X_j$ when the effects of the other inputs have been canceled.

If rank is set to true, then the rank coefficients are also calculated.

GSA examples

Lotka-Volterra Global Sensitivities

Let's run GSA on the Lotka-Volterra model to study the sensitivity of the maximum of predator population and the average prey population.

using DiffEqSensitivity, Statistics, OrdinaryDiffEq #load packages

First let's define our model:

function f(du,u,p,t)
  du[1] = p[1]*u[1] - p[2]*u[1]*u[2] #prey
  du[2] = -p[3]*u[2] + p[4]*u[1]*u[2] #predator
u0 = [1.0;1.0]
tspan = (0.0,10.0)
p = [1.5,1.0,3.0,1.0]
prob = ODEProblem(f,u0,tspan,p)
t = collect(range(0, stop=10, length=200))

Now let's create a function that takes in a parameter set and calculates the maximum of the predator population and the average of the prey population for those parameter values. To do this, we will make use of the remake function which creates a new ODEProblem, and use the p keyword argument to set the new parameters:

f1 = function (p)
  prob1 = remake(prob;p=p)
  sol = solve(prob1,Tsit5();saveat=t)
  [mean(sol[1,:]), maximum(sol[2,:])]

Now let's perform a Morris global sensitivity analysis on this model. We specify that the parameter range is [1,5] for each of the parameters, and thus call:

m = gsa(f1,Morris(total_num_trajectory=1000,num_trajectory=150),[[1,5],[1,5],[1,5],[1,5]])

Let's get the means and variances from the MorrisResult struct.

2×2 Array{Float64,2}:
 0.474053  0.114922
 1.38542   5.26094

2×2 Array{Float64,2}:
 0.208271    0.0317397
 3.07475   118.103    

Let's plot the result

scatter(m.means[1,:], m.variances[1,:],series_annotations=[:a,:b,:c,:d],color=:gray)
scatter(m.means[2,:], m.variances[2,:],series_annotations=[:a,:b,:c,:d],color=:gray)

For the Sobol method we can similarly do:

m = gsa(f1,Sobol(),[[1,5],[1,5],[1,5],[1,5]],N=1000)

Design Matrices

For the Sobol Method, we can have more control over the sampled points by generating design matrices. Doing it in this manner lets us directly specify a quasi-Monte Carlo sampling method for the parameter space. Here we use QuasiMonteCarlo.jl to generate the design matrices as follows:

N = 10000
lb = [1.0, 1.0, 1.0, 1.0]
ub = [5.0, 5.0, 5.0, 5.0]
sampler = SobolSample()
A,B = QuasiMonteCarlo.generate_design_matrices(N,lb,ub,sampler)

and now we tell it to calculate the Sobol indices on these designs:

sobol_result = gsa(f1,Sobol(),A,B)

We plot the first order and total order Sobol Indices for the parameters (a and b).

p1 = bar(["a","b","c","d"],sobol_result.ST[1,:],title="Total Order Indices prey",legend=false)
p2 = bar(["a","b","c","d"],sobol_result.S1[1,:],title="First Order Indices prey",legend=false)
p1_ = bar(["a","b","c","d"],sobol_result.ST[2,:],title="Total Order Indices predator",legend=false)
p2_ = bar(["a","b","c","d"],sobol_result.S1[2,:],title="First Order Indices predator",legend=false)


Parallelized GSA Example

In all of the previous examples, f(p) was calculated serially. However, we can parallelize our computations by using the batch interface. In the batch interface, each column p[:,i] is a set of parameters, and we output a column for each set of parameters. Here we showcase using the Ensemble Interface to use EnsembleGPUArray to perform automatic multithreaded-parallelization of the ODE solves.

f1 = function (p)
  prob_func(prob,i,repeat) = remake(prob;p=p[:,i])
  ensemble_prob = EnsembleProblem(prob,prob_func=prob_func)
  sol = solve(ensemble_prob,Tsit5(),EnsembleThreads();saveat=t,trajectories=size(p,2))
  # Now sol[i] is the solution for the ith set of parameters
  out = zeros(2,size(p,2))
  for i in 1:size(p,2)
    out[1,i] = mean(sol[i][1,:])
    out[2,i] = maximum(sol[i][2,:])

And now to do the parallelized calls we simply add the batch=true keyword argument:

sobol_result = gsa(f1,Sobol(),A,B,batch=true)

This user-side parallelism thus allows you to take control, and thus for example you can use DiffEqGPU.jl for automated GPU-parallelism of the ODE-based global sensitivity analysis!