Information visualisation and visual analytics Week 2 笔记

Information visualisation and visual analytics

Week2 笔记

1.      Information visualisation: 强调effectiveness,不只是makingbeautiful pics

 

 

2.      要思考的问题:

(1)   Data: what data users see? Data types

(2)   Tasks: why users tend to use a vis tool?

(3)   Idiom: how the visual encoding and interactionidioms are constructed in terms of design choices?

Data-tasks-idiom 合一起是一个instance

 

3.      设计vis tool时必须思考的三个问题：

(1)   Computational capacity, big data set?

(2)   Human perceptual and cognitive capacity

(3)   Display capacity, information density in asingle image

 

 

4.      Visualisation pipeline:

（1）     两个任务：

a.      初级：帮助理解你的task

b.      中级：帮助设计者refine, debug, extent: systems’algorithms, how the algorithms are affected by parameters

（2）      过程：（这不是一个完全线性的过程）

在visual representation时要思考需要用到什么技术

（摘自Uta Hinrichs, CS5044 InformationVisualisation and visual analytics, University of St Andrews）

 

5.      电脑能完成的工作是machine learning 范畴的。Vis 主要用来探求未知关系（needfor human judgement）。有很多possible question to ask, 需要powerfulpattern detection properties of the human visual system

 

6.      Vis tools 有长期和短期，短期通常在decision-making后就弃了，也可以用于Pre

 

 

7.      为什么可视化需要电脑呢？have a computer-based toolgenerated the visual representation automatically obviously saves human effortcompared to manual creation

 

8.      External data representation(XDR) 外部数据表示法，在OSI模型表示层presentationlayer中实现

 

 

9.      External representation/memory有许多形式：touchable,2D…

 

 

10.  Vis tools需要帮人们看清data structure，而不只是datasummary (在summary中很容易loseinformation)

 

 

11.  Interactivity 在handlecomplexity中起crucial作用

 

 

12.   Visualencoding

在个体接受外界信息时，对外界信息的视觉刺激进行编码。70年代波斯纳（美）等人利用减法反应时基本范式，在其实验中清楚地表明，某些短时记忆信息可以有视觉编码和听觉编码两个连续的阶段，这是认知心理学上的重大发现（AA与Aa）

 

13.  一个设计不effective的原因可能是：

（1）     Fail to match the features of the humanperceptual and cognitive systems

（2）      Badmatch to intended tasks

 

14.  Data representation

(1)   What can be visualised (datasets):

a.       Tables (basic datasettype)  (basicdataset type)

  i.           Flat tables: each row represents an item ofdata, and each column is an attribute of the dataset.

 ii.           Each column: attribute of the dataset

 iii.           Each row: item of data

 iv.           Each cell specifies the value of an item/attributepair

b.       Networks ((basic datasettype) can be presented through adjacency tables)

 i.           Specify relationships between two or multipleitems

ii.           Node (vertex): item in a network that can haveattributes (item 通常在network里叫node)

iii.           Link (edge): relationship between twoitems(nodes), can have attributes (different from the nodes’ attributes)

c.      Trees: networks with hierarchical structures.Trees 每个child node only has one parent node

d.      Fields (basic dataset type) (人体组织结构，理论无穷数量，x-ay)

e.      geometry ((basic dataset type) specifies the shape of items)

f.       Datasets的其他类型：trees, clusters (grouping-basedattribute similarity), sets (a set of unordered items), lists (array in cs). Complexcombinations of different dataset types are common. 通常一个dataset中会有几种data types

(2)   Static & dynamic datasets:

Static: items are available all at once, itemsdon’t change

Dynamic: items change over time (e.g.temperature data), new items are added over time

(3)   Metadata: data和metadata的界限并不是很分明。Metadata就是additionalinformation of the dataset, e.g. semantics

(4)   分析数据时我们需要知道它的semantics(its real worldmeaning)和type(包含data’sstructural/mathematical interpretation和data level (这些datatypes是如何组成一个largestructure的))

(5)   Data types:

a.      Grids (basic data type)

Specifiesthe strategy for sampling continuous data in terms of both geometric and topologicalrelationships between its cells.

b.     Positions (basic data type)

Spatialdata, providing a location in 2D or 3D space

c.      Items (basic data type)

People,stocks, cities: individual discrete entity: data point

d.     Attributes (basic data type)

Salary,price, number of sales

Propertyof an item that can be specified or measured

i.            Attribute types (很多时候data没有固定的attributetype, 在于我们怎么定义和使用data):

a)      Categorical (only distinguish whether two thingsare the same or different):

1)     Nominal

Nointrinsic ordering, operations只有等于和不等，比如电影类别，城市名称

2)     Interval

3)     Hierarchical: parent-to-child, 只有一个爸

Time:hour, day, week, month, year

b)     Ordered (can be sequential, diverging, cyclic):

1)     Ordinal

Rankingis possible BUT mathematical operations do not make sense

2)     Quantitative: math operations make sense

e.      Links (basic data type)

Relationshipbetween items, typically within a network

Typically,a visualisation aims to highlight particular relationships between items and/orattributes within or across data sets.

(6)   Dataset和data type的三个示例

Fields	geometry	Clusters, sets, lists
Grids, positions, attributes	Items, positions	items

 

(7)   Key v.s. value semantics

A key attribute can be independent attributes(in statistics) or dimension (in DB).

Value attributes can be dependent attributes (instats) or measure (in DB).

A key attribute acts as an index that is used tolook up value attributes.

一个simple flat table 只有一个Key（比如ID）， whereeach item corresponds to a row in the table, and any number of valueattributes.

(8)   Marks (data-driven): visual mapping

a.      Point 2D或3D

b.      Line 1D

c.      Area 2D

在table dataset中，一个Mark一般代表一个item/node/link

The two link mark types are:

a.      Connection: a pairwise relationship between twoitems, using a line

b.      Containment: shows hierarchical relationshipusing areas

Connect marks can be nested within each other atmultiple levels

对marks的使用应遵守theprinciples of expressiveness and effectiveness

(9)   Visual variables (channels): visual mapping【channels的选择会极大影响可视化效果】

a.      Perceptual characteristics（注意区分selective和associative）

i.                    Length: how many distinctions can we perceive?

ii.                  Selective: can marks be perceived as different?

iii.                Associative: can marks be perceived as similar?

iv.                Quantitative: can marks be perceived asproportional to each other?

v.                  Order: can an order in marks be perceived?

b.     Variables:

i.                    Position (core 5 to consider)

Willdominate the users’ mental model心智模型, 五个perceptualcharacteristics 都能做到

ii.                  Size (包括area) (core 5 to consider)

五个perceptualcharacteristics 都能做到，但是都很有限

iii.                Shape (core 5 to consider)

做不到quantitative和order, shape用在associate里很多

iv.                Color (value) (core 5 to consider)

禁止用在quantitative里

v.                  Color (hue色彩，色度) (core 5 to consider)

这是最复杂也最powerful的part

禁止用在quantitative和order(hardto get common senses)里

vi.                其他visual variables: orientation,texture, saturation, transparency, curvature, movement, flickr网路相簿

c.      Channel 分两种：

i.        Identity channels tell us information about whatsomething is or where it is

ii.      Magnitude channels tell us how much of somethingthere is

d.      Ordered attributes should be shown with themagnitude channels: aligned spatial position/unaligned spatial position,length, angle, area, depth, luminance(亮度), saturation, curvature, volume(3D size) 【否则违反expressiveness原则】

e.      Categorical attributes should be shown with theidentity channels: spatial region, hue, motion, shape 【否则违反expressiveness原则】

f.       如果不想排序，就不要随便用亮度来做可视化

15.  Using marks and channels的两大原则：

(1)   Expressiveness (新手常见错误)

The visual encoding should express all of, andonly, the information in the dataset attributes

对于排序数据，应该在展示数据时做到让我们的感知perceptualsystem intrinsically senses as ordered

同理，对于categorical data， 不应该出现suggest ordering的情况

(2)   Efficiency

The importance of the data attribute should beencoded with the most effective visual variable, in order to be noticeable

i.                    Discriminability辨别力

Selective和 associative

Thecharacterisation of visual channel thus should quantify the number of bins thatare available for use within a visual channel, where each bin is adistinguishable step or level from the other.

ii.                  Accuracy

Psychophysics精神物理学问题 s=Iⁿ , the psychophysics power law of stevens. S是感知sensation，I是physicalintensity, n 是the power law exponent [0.5, 3.5] thesublinear 0.5 for brightness to the super linear 3.5 for electric current. Thesublinear phenomena are compressed.

图摘自Tamara Munzner, Visualisationanalysis & design

iii.                Separability可分离性: 很难把所有visualvariables独立对待，因为有一些有dependencies andinteractions with others

Howdoes this visual varibales interact with other visual vaiables

(3)    

图摘自Tamara Munzner, Visualisationanalysis & design, 自带ranking

两种attributes position都是第一位的，所以设计的时候要确保自己没乱用position和color

16.  Relative v.s absolute judgements

人类的感知系统是基于相对判断的，而不是完全判断(Weber’s law)，所以一般要有比较才能有意义

17.  色彩恒常性

当照射物理表面的颜色发生变化时，我们对这个物体表面颜色的直觉依然会保持不变的知觉特性（其实是能感知到变化的，只是一定范围内认为不变，不敏感）。在图像处理科学领域中，给予人眼对景物的认知特点，把图像中背景元素与光照元素分离是解决问题的关键。