SepiaSensitivity

The SepiaSensitivity module implements Sobol-index based sensitivity analysis.

sepia.SepiaSensitivity.sensitivity(model, samples_dict=None, ngrid=21, varlist=None, jelist=None, rg=None, option='mean')

Compute sensitivity Sobol indices. (Warning: not fully tested for all model types and input choices.)

Parameters

  • model (sepia.SepiaModel) – instantiated SepiaModel with MCMC samples

  • samples_dict (dict/NoneType) – selected samples from model.get_samples(flat=True) (default: uses all samples in model)

  • ngrid (int) – number of grid points in each dimension for calculation of mian/joint effects (default: 21)

  • varlist (list/string/NoneType) – list of tuples giving pairs of variables for which joint effects are desired; using ‘all’ indicates to compute joint effects for all variables. Default is None.

  • jelist (list) – list of tuples indicating variables for which joint sensitivities are desired (similar to varlist; default is None)

  • rg (numpy.ndarray/NoneType) – matrix with one row for each variable giving min/max values for sensitivity calculations assuming unit hypercube scaling, shape (num_vars, 2), default: unit hypercube.

  • option (string/dict) – do calculations based on ‘mean’ (posterior mean GP params), ‘median’ (posterior median GP params), ‘samples’ (GP param samples in samples_dict), or pass dict of samples from model.get_samples(flat=True)

Returns dict
Depending on input options, sens dict may contain:

  • totalMean (overall output mean – posterior mean)

  • totalVar (total output variance – posterior samples)

  • smePm (main effect sensitivity indices – posterior mean)

  • stePm (total effect sensitivity indices – posterior mean)

  • siePm (two-factor ineraction effect sensitivity indices – posterior mean)

  • sjePm (joint effect sensitivity indices – posterior mean)

  • mef (main effect functions by basis component – posterior mean and SD)

  • tmef (main effect functions – posterior mean and SD)

  • tjef (two-factor joint effect functions – posterior mean and SD)

  • sa (dict with information by basis coefficient; keys, e0 - overall output mean, vt - total output variance, sme, ste, sie, sje, mef, jef)