Large Scale Data Analysis - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 330

Fig. 27.8 Our framework provides a natural and intuitive work-flow for the exploration of global
trends in feature-based statistics. The linked view system allows users to display susbsets of features
using range sliders and/or by selecting regions of CDFs and histograms. This image shows regions
of locally high scalar dissipation rate in a temporally-evolving turbulent CO/H 2 jet flame
p is used to select an initial set of features from the clan, which are then further
subselected using the subselections Q .
Species distributions plots include histograms and empirical CDFs, and track
the distribution of the attribute att i . A time-series, as the name suggests, shows the
evolution of att i over time, and requires an additional family-wide reduction operator,
R f , as input. Parameter studies are an extension of time-series that show how att i
changes as the parameter p is varied. For these plots a clan-wide reduction operator,
R c , is required in addition to R f . Note that parameter studies can be come expensive
as the range and granularity of p increases, because attributes must be aggregated
for each p -value independently. While parameter plots are the most expensive to
produce they are also often very useful. In particular, a parameter plot shows how
stable or unstable a given analysis is to the parameter selection. This is crucial in
any exploratory setting to guarantee that the basis of important conclusions is not an
inherently unstable analysis.
We provide a convenient GUI that allows the user to specify which attributes they
would like to explore, loading only those to minimize memory overhead. Subselec-
tion sliders are generated for each specified attribute automatically and, if multiple
hierarchies are available, the user can toggle between these and can update parame-
ters interactively. Optional log scaling is provided, and radio buttons are used for
selection of family- and clan-wide reduction operators. The plot viewer is linked to
the feature browser to provide context as statistics are explored. Only those features
that have been subselected using the GUI sliders are displayed by the feature browser.
Users can click on an individual feature in the feature browser to obtain details on
its associated statistics. Furthermore, when the user picks regions of histograms or
CDFs, only those features that are contained in the selected bins are displayed by the
feature browser, see Fig. 27.8 .