Fig. 3.

(A) The parameter sweep continues. Here a total of 1553 samples are being observed. Label propagation enables the modeler to change and add labels to points as the sweep continues. Here, the focus and thus the preferences has been directed to the upper part of the data blob. A new label propagation is performed yielding new probabilities of unknowns. (B) To simulate the process of a massive parameter sweep and the robustness of the current state of the system, we sample parameter points associated with robust oscillations. These points become mapped directly to our ROI. To reduce the uncertainty in the prediction model, the modeler can be queried to label points with high uncertainty using e.g. entropy based measures. (C) Zooming in on ROIs by neglecting points with low probability of belonging to the interesting class. We observe some non-robust outliers in the ROI. This suggest that we need other features than the minimal set to separate these from the robust oscillations