1

Stack Zooming forMulti-Focus Interaction in

Time-Series Data Visualization

Waqas Javed (wjaved@purdue.edu)Niklas Elmqvist (elm@purdue.edu)

Presented by Jean-Daniel Fekete

2

Motivation: Multi-Focus Interaction

Motivation

Mélange [Elmqvist 2008]

Multiple Focus Regions

3

Outline

Stack Zooming Introduction

Stack zooming in detail

Layout and correlation graphics

Stack zooming in action

The TraXplorer System System design

Visual interface

Video Demonstration

Summary

Future Work

4

Introduction

Time-series data tends to be long and often its analysis requires comparison across multiple focus regions

Current time-series visualization tools have limited support for comparing several foci while retaining context

Stack zooming is a method for supporting this kind of multi-focus interaction in time-series data exploration

Based on building hierarchies of stacked 1D strips Each subsequent stack represents a higher zoom level

Sibling strips represent branches in the visual exploration

5

Layout and Correlation Graphics Stack zooming is based on creating a stack of zoom areas

Nodes in a zoom stack are laid out on the visual substrate using a space-filling layout algorithm Splits the vertical space by the depth of the zoom stack

Splits the horizontal space by the number of siblings at each level

6

Layout and Correlation Graphics

Layout allocations can be changed by dragging the borders of a strip

The order of child strips for each level in the zoom stack is significant for conveying the positions of the displayed intervals of a time series

The layout manager will always order child strips for each level in the zoom stack to be the same as the order of their intervals on the parent strip

7

Layout and Correlation Graphics

Relationships between parent and child strips in adjacent levels of zoom stack must be visible Focus

Context

Distance awareness

We discuss three different correlation graphics that visually indicate the relationships between different visual strips in the zoom stack

8

Layout and Correlation Graphics Color-coded zoom areas:

Parent strips show color-coded semi-transparent selection areas

Indicates the position and extents of each child strip in the time series

Color-coded strip frames:

Child strips have color-coded frames that correspond to the color of its parent selection area

This gives a visual link between parent and child

9

Layout and Correlation Graphics

Color-coded zoom areas: Parent strips show color-coded

semi-transparent selection areas

Indicates the position and extents of each child strip in the time series

Color-coded strip frames: Child strips have color-coded

frames that correspond to the color of its parent selection area

This gives a visual link between parent and child

10

Layout and Correlation Graphics

Correlation links:

Explicit correlation links drawn as dotted lines and arrows from zoom areas in parents to the children

Allows for quickly understanding the correlation structure

May be shown in a transient overlay to minimize visual clutter

11

Stack Zooming in Action

When the user begins to analyze the dataset, the whole display is taken up by the full time series drawn as a line visualization on a single strip

12

Stack Zooming in Action

Using a drag on the surface of this strip, the user can create a child strip of the main strip that displays the selected subset of the time data

13

Stack Zooming in Action

Additional zoom operations on any of the dataset strips will create additional children in the zoom stack

14

The TraXplorer System

15

System Design

TraXplorer is designed to support a communication-minded iterative workflow Exploration

Collaboration within the analysis team

Dissemination to external stakeholders

16

The Visual Interface

Components: Main visualization window

Data box

Layer control box

Presentation tree window

17

The Main Visualization Window

The main visualization window is a visual space supporting stack zooming

Contains a visualizations of time-series data on a common time axis and potentially different value axes

Visualization type is independent of the layout management

Our implementation currently supports basic line graphs, filled line graphs, and horizon graphs

18

The Layer Control Box

Each data series is a unique layer in TraXplorer

The layer control box can be used to move, to delete, and to toggle the visibility of individual tracks, as well as to change color mapping, transparency, and track title

Used to determine which track should be used for the value axis labels

1919

The Layer Control Box

Two or several tracks can be linked to use the same scale for the value (Y) axis, thereby supporting direct comparison of values

20

The Data Box

The data box displays local statistics about the currently selected region

Detail-on-demand for computing measures for a particular track Min/max, average, median, standard

deviation, etc

Add comments to any particular track

Checkboxes to add this data to the visual display of the track

21

The Presentation Tree

The presentation tree is a hierarchical representation of the zoom stack

The analyst can prune, move, or hide individual zoom nodes (i.e. child strips) using the presentation tree to refine the presentation

Can access the exploration history using the presentation tree to linearize the combined exploration sessions of the data similar to a slideshow presentation suitable for presentation to the audience

22

Video

23

Summary

Theoretical background of a novel multi-focus interaction technique called stack zooming Multiple focus points in time-series dataset visualizations

Context, distance, and relationships between time-series

The TraXplorer implementation A visual interface to support the time series exploration

Supports stack zooming

Communication-minded workflow

24

Future Work

Study the empirical performance of the tool in comparison to similar tools

Improve the tool to better support collaborative visual exploration settings involving teams of analysts working together

Study how the tool can help analysts fill different roles in the analysis process

25

Questions?

Thanks!

Contact information: Waqas Javedwjaved@purdue.edu

Merci JD!!!

http://web.ics.purdue.edu/~wjaved/projects/stackzooming