Composition with granularity operators

This combination realises one of the aims of providing different perspectives on the data: the hierarchical structure can be modified to reflect multiple temporal scales enabling encodings to match the characteristic of these scales and tasks related to these scales. Within the category of granularity operators, binning and granularity changing operators are suitable for this approach. The use of rotate and align operators may be necessary to adjust encodings after transformations.

Bin/Change Granularity + Juxtapose Between Levels

Signature: (G,E1) Ñ J ((G,E1),(H,E2))

Changing the level of aggregation while juxtaposing (fig. 7.8) allows different levels to be compared in separate views. Juxtaposing while drilling down or rolling up time allows the visual ordering of levels to reflect the granularity hierarchy, whereas changing to specific granularities may not have the same effect.

Bin/Change Granularity + Superimpose Between Levels

Signature: (G,E1) Ñ S ((G,E1),(H,E2))

This combination allows the use of visual marks to be layered over each other, enabling multiple scales to be compared on the same view. In comparison with juxtaposing, superimposition allows a larger rendering area to be used, although with the risk of increasing clutter and occlusion. Figure 7.9 shows an example where lines are drawn over a bar after the granularity has been changed; the bar encodes the average value for the time point, which is expanded into two time points in the line chart.

Bin/Change Granularity + Nesting Between Levels

Signature: (G,E1) Ñ N ((G,E2),(H,E1))

This last configuration is the basis for calendar and grid-based visualisations. The granularity of an unmodified linear time domain can be changed, with each level being

nested within each other. Figure 7.10 shows how successive applications of the drill

down operator along with adding a new nested level to the hierarchy results in a grid arrangement.

(a) (G, E1) (b) (c) J ((G, E1),(H, E2)) (d)

Fig. 7.8 Example of changing granularity and juxtaposing, with the colours of the shapes matching the nodes in the trees. In the second row, the tree has been drilled

down to a finer granularity H, which is encoded as circles connected by a path and juxtaposed below the first level.

7.4

Chapter summary

This chapter described how the components are connected with each other under the unified framework. The effects of the temporal transformations on the view component were analysed; for the combination of the transformations and composition methods, it was explained how the interplay between the two components empowers the visual exploration driven by the various transformations of time. Various examples were given demonstrating the visual result of the changes in the hierarchical structure as a result

(a) (G, E1) (b) (c) S ((G, E1),(H, E2)) (d)

Fig. 7.9 Example of changing granularity and superimposing, with the colours of the shapes matching the nodes in the trees. In the second row, the tree has been drilled

down to a finer granularity H, which is encoded as circles connected by a path and

superimposed over the first level.

of the transformations. The next chapter describes in further detail the paths that can be explored in the design space that emerges from the framework.

(a) (G, E1) (b) (c) N((G, E1),(H, E1)) (d) (e) N(N((G, E1),(H, E1)),(I, E2)) (f)

Fig. 7.10 Example of changing granularity and nesting, with the colours of the shapes matching the nodes in the trees. In the second row, the tree has been drilled down to a finer granularity H, with the corresponding view repeating the same encoding. In the third row, the tree has been drilled down to a finer granularity I ; this time, the corresponding view uses encoding E2 with the orientation of the spines changed to

Case study

This chapter demonstrates how the framework supports exploring temporal data through hierarchical visualisations. Questions and data from a visualisation challenge are used to explain how the different aspects of the framework support the design of

multiple perspectives, as part of an EDA approach to solve the challenge. The aim of

this chapter is not to discuss the best configurations for the tasks, but demonstrating the design process and the possibilities that the framework enables regarding its three components.

8.1

The dataset

The VAST Challenge1 _{is a traditional annual data and visualisation challenge, in which}

participants design interactive visualisation tools for use in a fictional scenario inspired by real-world datasets and tasks, including fictional locations, characters and data. As part of the 2017 challenge, the setting involved the analysis of temporal patterns for vehicles that roam through a nature reserve with electronic gates that record their passage. Vehicles are assigned ID’s and a vehicle type; the gates include exit and entrance gates from the reserve and other types of gates, including camping areas, locations where park rangers scout and other vehicles are not allowed to pass through and the park rangers headquarters. The dataset contains timestamped records of the vehicles – the challenge is to design visualisations that help solve challenge. The list of attributes for each record is as follows:

• Timestamp: including year, month, day, hour, minutes and seconds granulari- ties;

• Car-id: a unique car identifier that is assigned to every vehicle that enters the reserve and is reused for subsequent visits;

• Car-type: classification of vehicles into 7 numeric identifiers: 1 for car or motorcycle, 2 for two-axle trucks, 3 for three-axle trucks, 4 for four-axle and above trucks, 5 for for two-axle buses and 6 for three-axle buses. Additionally, park rangers vehicles have a P suffix; only two two-axle trucks are used, as such, park ranger vehicles are identified as 2P. A general identifier car-category can be extracted from this variable, with a category encompassing every type of truck or bus, for example.

• Gate-name: identifiers for the following gate types: entrance, gate, general-gate,

ranger-stop, camping and ranger-base. Each unique gate is given a numeric suffix,

resulting in entrance0, general-gate1, etc. A general identifier gate-type can be derived from this variable by removing the numeric suffix.

The following three questions that are asked in the challenge are suitable for an exploratory approach and will be investigated in this chapter: (1) describing daily patterns of vehicles, including where they happen in the park, when they happen during the day and a hypothesis of what the pattern means (for example, vehicles simply passing through the reserve); (2) describing multi-scale patterns of vehicles, such as periodic visits by groups of vehicles at the same location and (3) describing unusual patterns, such as vehicles passing through gates that they are not allowed or patterns that are difficult to hypothesise given expected behaviour of vehicles, such as long stop-overs in the reserve without visiting a campsite. Due to the size of the dataset, throughout the chapter subsets of the data will be selected for the visualisations. This leaves out important visualisation and interaction decisions about reducing the number of items to be visualised; the aim of the chapter is demonstrating the use of the framework for the tasks rather than designing an interactive application that would help solve the challenge. The author has not participated in the challenge.