"Global sensitivity indices for nonlinear mathematical Design and estimatorįor the total sensitivity index." Computer Physics Communications, 181(2):259-270, "Variance based sensitivity analysis of model output. European Journal of Operational Research, 226(3), 536-550. Global sensitivity measuresįrom given data. (2017) SALib: An open-source Python library for sensitivityĪnalysis. The `N_override ` keyword argument allows users to override the `N ` used in a specific `analyze ` call to analyze just a subset (useful for convergence graphs).Īn example of the basic flow can be found in src/main.jl using the Ishigami test function in src/test_functions/ishigami.jl, and is copied and commented below for convenience.īorgonovo, E. The `progress_meter ` keyword argument indicates whether a progress meter will be displayed and defaults to true. If these are Nothing than no confidence intervals will be calculated. Parameters defined using the `num_resamples ` and `conf_level ` keyword args. Performs a Sobol Analysis on the `model_output ` produced with the problem defined by the information in `data ` and returns the a dictionary of results with the sensitivity indices and respective confidence intervals for each of the The signature for this function is as follows.įunction analyze(data :: SobolData, model_output :: AbstractArray = nothing) Sampling with sample is the first of the two main steps in an analysis, generating the model inputs to be run through a model of choice and produce the outputs analyzed in the analyze function. Note: For now the sample function will call the most used sampling protocol for the particular method, (Sobol sequence for Sobol method and Latin Hypercube sampling for Delta method), but in this future this will be rearranged and generalized since, for example, the Delta method can also just as well use Sobol sequence sampling and other methods. These two functions call methods based on the type parameterization of their data argument, which is either of type SobolData or DeltaData. The API contains two primary functions: sample and analyze. 2010)ĭelta Moment-Independent Measure ( Borgonovo 2007, Plischke et al. Sobol Sensitivity Analysis ( Sobol 2001, Saltelli 2002, Saltelli et al. The package currently includes the following methods: The pacakge seeks to implement several of these same algorithms in Julia along with providing a clear, user-friendly API. Much of this package is based on SALib (Herman and Usher, 2017) which implements several global sensitivity analysis measures in Python. A Julia package which implements global sensitivity analysis methods.