Qualitative sensitivity analysis

econsa applies the methods in [4] to calculate morris indices for models with dependent parameters. The Elementary Effects (EE), also known as the Morris method, is a qualitative way to screen inputs and helps to determine the set of influential and non-influential inputs. Shapely values on the other hand, …

Constributor: Janos Gabler (janosg)

Calculate morris indices for models with dependent parameters.

We convert frequently between iid uniform, iid standard normal and multivariate normal variables. To not get confused, we use the following naming conventions:

-u refers to to uniform variables -z refers to standard normal variables -x refers to multivariate normal variables.

econsa.morris.elementary_effects(func, params, cov, n_draws, sampling_scheme='sobol', n_cores=1)[source]

Calculate Morris Indices of a model described by func.

The distribution of the parameters is assumed to be multivariate normal, with mean params["value"] and covariance matrix cov.

The algorithm is based on Ge and Menendez, 2017, (GM17): Extending Morris method for qualitative global sensitivity analysis of models with dependent inputs.

Parameters
  • func (function) – Function that maps parameters into a quantity of interest.

  • params (pd.DataFrame) – DataFrame with arbitrary index. There must be a column called value that contains the mean of the parameter distribution.

  • cov (pd.DataFrame) – Both the index and the columns are the same as the index of params. The covariance matrix of the parameter distribution.

  • n_draws (int) – Number of draws

  • sampling_scheme (str) – One of [“sobol”, “random”]. Default: “sobol”

Returns

  • mu_ind (float) – Absolute mean of independent part of elementary effects

  • sigma_ind (float) – Standard deviation of independent part of elementary effects