SensitivityAnalysis.run_sweep

SensitivityAnalysis.run_sweep(var_input, sweep_values, *, var_names='channel_contribution', sweep_type='multiplicative', posterior_sample_fraction=1.0, response_mask=None, extend_idata=False, **kwargs)

Run sweeps by forward-evaluating the response graph (no Jacobian).

Parameters:

var_inputstr

Name of the pm.Data variable (e.g., “channel_data”). Expected shape: (date, *dims, arbitrary_dim) that match var_input dims.

sweep_valuesnp.ndarray

Values to sweep over.

var_namesstr

The deterministic variable of interest.

sweep_type{“multiplicative”,”additive”,”absolute”}

Type of sweep to apply.

posterior_sample_fractionfloat

Posterior sampling control in (0, 1]. The retained fraction of draws is roughly posterior_sample_fraction ** 2 (e.g., 0.1 keeps about 1% of the draws, while 1.0 keeps the full posterior).

response_maskxr.DataArray | np.ndarray | None

Optional boolean mask with the same non-sample dims as var_names. When provided, the mask zeroes out the response prior to evaluating the sweeps. Useful for focusing on specific coordinates of the deterministic response (e.g., single channels).

Returns:

xarray.DataArray | None

If extend_idata is False, returns an xarray.DataArray with shape (sample, sweep, *dims_order), where dims_order are the non-date dims of var_input in the same order as in the model. The response is averaged over the date axis as in the draft example. If extend_idata is True, stores the result under idata.sensitivity_analysis and returns None.