
Review: Validation

Where are We?

The Big Picture
- Datasets are often large and complex so drawing all of them into a single static view can be overwhelming.
- What options are available to handle such complexity? There are five.
- 1. Change a view over time (this lecture)
-
- Derive new data (Lecture 3)
- 3. Facet data by partitioning into multiple juxtaposed views (this lecture)
-
- Reduce the amount of data (next lecture)
-
- Embed focus and context information within a single view (next lecture)
Change View over Time
- Changing the view over time is the most obvious, popular, and flexible choice in vis design.
- Unavailable in printed visualizations
- Let users modify the view.
- Change the encoding
- Change the arrangement
- Change the order
- Change the viewpoint
- Change which attributes are filtered
- Change the aggregation level
- $\dots$
Changing the Encoding
- Tableau
- Encoding
- Arrangement
- Mark type/size/color
- $\dots$
Changing the Order
- Reordering, often called sorting, is a powerful choice for finding patterns in a dataset by interactively changing the attribute that is used to order the data.
- Widely-used for table datasets
- Order by student_id, order by score, …
- What if there are multiple attributes that can be used for sorting?
Example: LineUp
- Designed to support exploration of tables with many attributes through interactive reordering and realigning.

- Order universities by 0.25 * (Research) + 0.25 * (Citations) + $\dots$
- Changing the alignment and arrangement of stacked bars:

Summary: LineUp

Animated Transition
- When changing a view, you can provide animated transition.
- A series of frames is generated to smoothly transition from one state to another.
- vs. jump cut (no animation, simply redraw the view)
- https://sanddance.js.org/app/


- Benefits:
- Help users maintain a sense of context
- Animation seems cool
- Drawbacks:
- Hard to implement
- Hard to follow if too much change
- Change blindness
- cf. staggered animation

Staggered Animation
- 1) Color transition + move up
- 2) Move left

Select an Element
- Users need to select an element before they change it.
- Design choices:
- What can be a target?
- a bar? a group of bars? an axis? a view? a legend?
- What are the possible selection types?
- selected / not selected
- selected + hovered / selected / hovered / not selected
- {selected, hovered, faded out}!
- How many items can be in the selection at once? one? many?
- What is the default? empty? a default item? all?
Highlight an Element
- Selected elements are highlighted by changing their visual appearance, providing users with immediate visual feedback.
- Design choices for highlighting
- Changing color
- Adding an outline
- Using the size channel
- Using motion coding
- Adding shadows

Example: Context-Preserving Visual Links
- Links are drawn as curves that are carefully routed between existing elements in a vis.

Navigate a View
- Navigation: changing the point of view from which things are drawn
- Only items in the frame are visible
- Three components of navigation:
- Zooming: moves the camera closer to or farther from the plane
- Panning: moves the camera parallel to the plane of the image, either up and down or from side to side
- Rotating: spins the camera around an axis
Zooming
- Geometric zooming: analogous to navigation through the physical world
- Things get bigger but not change.
- Semantic zooming: the representation of the object adapts to the number of pixels available in the image-space region occupied by the object
- Things get bigger and change.

Reducing Attributes
- Given 3D data points,
- A slice operation leaves data points that have a specific value on a specific dimension.
- (x, y, z) such that z = 3.0
- Dimensionality is reduced by 1
- A cut operation leaves data points under or over a cutting plane
- (x, y, z) such that ax + by + cz < d
- A project operation multiplies a projection matrix X (linear projection).
Example: Slice and Cut

Example: Project
- Principal Component Analysis (PCA)

Facet into Multiple Views
- Facet: split data into multiple views
- Design choices:
- Do the two views use the same visual encoding?
- Do the two views share data?
- Do the two views share navigation?
- If we pan one view, does the other change?
- How are the two views placed?
Coordinated Views
- Coordinated/linked views: multiple views relevant to each other
- Do the views use the same encoding?
- If yes, shared encoding
- Otherwise, multiform encoding
- How much data is shared between the coordinated views?
- Shared data if all data is shared.
- Overview/detail if one view shows a subset of what is in the other.
- Small multiples if the views show different partitions of the dataset into disjoint pieces.

Example: Exploratory Data Visualizer
- If multiform views are coordinated, we can inspect multiple aspects of the same data at once (with different encodings)
- Called linked highlighting (very common interaction idiom in practice)

Summary: Exploratory Data Visualizer

Overview/Detail
- In overview/detail, one of the views shows information about the entire dataset to provide an overview of everything.
- e.g., minimaps

Example: Bird’s-Eye Maps



Small Multiples
- In small multiples, views share the same encoding but show different data (different set of rows)

Example: Cerebral

Summary: Cerebral

Shared Navigation
- In many overview/detail approaches, navigation is synchronized between the overview and the detail view

Partition into Views
- A multiattribute dataset should be partitioned into meaningful groups to be visualized in multiple views.
- = How to create multiples in small multiples?
- How to partition?
- by a categorical attribute (e.g., grouping)
- by a quantitative attribute (e.g., binning)
- by similarity (e.g., clustering)
- How to align partitions?
- List alignment, matrix alignment, and recursive subdivision

Example: Grouped Bar Charts (List Alignment)

Example: Trellis (Matrix Alignment)
- Group by site
- 6 levels
- where the barley was grown
- Group by year
- Map variety to y
- Map yield to x
- Main-effect ordering in (a)

Example: Trellis
- Color encoding for comparison

Example: HiVE (Recursive Subdivision)
- Visualizing one million property transaction data in London
- House type: Flat (flats), Ter (attached terrace houses), Semi (semidetached house), and Det (fully detached house)
- Time of sale: year/month
- Neighborhood: 33 levels
- How would you visualize this data?
- Split by house type
- Split by neighborhood
- A cell is a heatmap that shows time with years as columns and months rows.

- Split by neighborhood first
- Split by house type
- A cell is a heatmap that shows time with years as columns and months rows.

- Split by house type (Treemap)
- Split by neighborhood
- A cell is a heatmap that shows time with years as columns and months rows.

- Split by house type
- Arrange by neighborhood

Superimpose Layers
- Superimposed layers: you can place multiple views at the same location
- Design choices:
- How many layers are used?
- How are the layers visually distinguished from each other?
- Is there a small static set of layers that do not change, or are the layers constructed dynamically in response to user selection?
Example: Static Layers

Example: Superimposed Line Charts
- Shows the scalability issue of superimposition

Example: Horizon Graph

Example: Dynamic Layers
- With dynamic layers, a layer with different salience than the rest of the view is constructed interactively.
- Highlight one-hop-away nodes

Summary – Interacting with Views
- Changing a single view (manipulate): encoding, arrangement, order
- Viewpoint: pan, zoom (geometric/semantic), rotate
- Data: slice, cut, and project
- Multiple coordinated views (facet)
- Juxtaposition (multiform, overview/detail, small multiples)
- Partition
- Superimpose (static/dynamic)

댓글남기기