variance_based_decomp

Computes Sobol’ main effects using a binned approach

Specification

  • Alias: None

  • Arguments: None

  • Default: no variance-based decomposition

Child Keywords:

Required/Optional

Description of Group

Dakota Keyword

Dakota Keyword Description

Optional

drop_tolerance

Suppresses output of sensitivity indices with values lower than this tolerance

Optional

num_bins

Number of bins used to compute the variance-based decomposition

Description

Uses unstructured input-output samples to estimate main effect indices. It cannot compute total indices.

Expected Output Sensitivity indices for main effects only will be reported. Main effects (roughly) represent the percent contribution of each individual variable to the variance in the model response.

Examples

method,
  sampling
    sample_type lhs
    samples = 100
    variance_based_decomp
      vbd_sampling_method binned

method,
    import_points
        import_points_file "all_samples.dat"
        variance_based_decomp

Note that since the import_points method only supports the binned approach, vbd_sampling_method is not specified.

Theory

The binned approach to computing Sobol’ main effect indices is introduced in [LM16]. As opposed to pick-and-freeze approaches like [STCR04], it does not require a specific sampling structure. Given a set of randomly-generated input-output samples, it computes the main effect index by binning samples, computing a sample statistic for each bin, then computing another sample statistic over the bins.

Two algorithms are detailed in [LM16]: computing a sample expectation for each bin, then a sample variance, or computing a sample variance for each bin, then an expectation. The second algorithm is implemented in Dakota.