Data Visualization, CS-171, Harvard Flashcards

Convey graphical information derived from data. It is based on exploiting the human visual system as a means of communication because it is a very high-bandwidth channel to the brain.

NOTE: Visualizations USED TO refer to mental images that people formed while they thought...

(1) To record information / have a graphical record of something; (2) to analyze data, reveal trends and patterns, and support reasoning; (3) confirm a hypothesis about the data; (4) communicate ideas / persuade / convince / inspire others.

In 1967 by Doug Engelbart at Augmentation Research Center (ARC). In 1968 he performed what is now called The Mother of all Demos where he showed off the mouse.

We're in the "revolution of industrial data" and data collection has never been easier--and will continue to get easier. We have RFID sensors, road sensors, live cameras, traffic sensors, chips everywhere and it will only increase.

We just have too much data. For example, a telescope used to just give a view of the skies. But now telescopes collect terabytes and petabytes of data that needs analysis.

“The ability to take data—to be able to understand it, to process it, to extract value from it, to visualize it, to communicate it—that’s going to be a hugely important skill in the next decades... because now we really do have essentially free and ubiquitous data.”

We have limits of cognition; we are easily distracted; we are already multitasking and can only absorb so much; we are forgetful and our working memory only has so much active capacity; we have attentional blindness; etc. Visualization helps us keep up.

Helps us think; uses perception to offload cognition; converts static data into useful information; serves as an external aid (e.g. when we record and store) to augment working memory; allows us to see vast amounts of information within our limited field of view; accelerates search and cognition; boosts our cognitive abilities.

“Visualization is really about external cognition, that is, how resources outside the mind can be used to boost the cognitive capabilities of the mind.”

Almost half--clearly dominant

Professor Hanspeter Pfister

When text and images are related they should be placed in close proximity--and to avoid using, "See Figure X on page..."

We see very little at any given instant, but we can sample any part of our visual environment so rapidly with swift eye movement, that we think we have all of it at once in our consciousness experience. We get what we need, when we need it.

The brain, like all biological systems, has become optimized over millennia of evolution. Brains have a very high level of energy consumption and must
be kept as small as possible, or our heads would topple us over. Keeping a copy of the world in our brains would be a huge waste of cognitive resources and completely unnecessary. It is much more efficient to have rapid access to the actual world—to see only what we attend to and only attend to what we need—for the task at hand.

Very little. Seeing is all about attention. This new understanding leads to a revision of our thinking about the nature of visual consciousness. It is more accurate to say that we are conscious of the field of information to which we have rapid access rather than that we are immediately conscious of the world.

Visual thinking consists of a series of acts of attention, driving eye movements and tuning our pattern-finding circuits. These acts of attention are called visual queries.

Determine a trend from a stock chart or try to get from point A to B using a map.

We can resolve about 100 points on the head of a pin held at arm's length in the very center of the visual field called the fovea. (The fovea is like a high-resolution lense.) Over half of our visual processing power is concentrated in a slightly larger area called the parafovea.

The non-uniformity of the visual processing power is such that half our visual brain power is directed to processing less than 5 percent of the visual world. This is why we have to move our eyes; it is the only way we can get all that brain power directed where it will be most useful. Non-uniformity is also one of the key pieces of evidence showing that we do not comprehend the world all at once. We cannot possibly grasp it all at once since our nervous systems only process details in a tiny location at any one instant.

Rapid movement of both eyes (saccade) which direct the fovea at interesting and useful location, pausing briefly at each, before flicking to the next point of interest.

The blind spot is an area of the retina where the optic nerve and blood vessels enter the eye. The brain works around this which is more evidence that seeing is not all the passive registration of information. Rather, it is active and constructive.

Broadly speaking, the act of perception is determined by two kinds of processes: bottom-up , driven by the visual information in the pattern of light falling on the retina, and top-down , driven by the demands of attention, which in turn are determined by the needs of the tasks.

About three.

Because visual working memory only holds about three objects which can be displaced by a more task-relevant action. This illustration shows how the binding of concepts that are "activated" (such as seeing a trading signal) and knowledge occurs. Th is momentary binding together of visual information with nonvisual concepts and action priming is central to what it means to perceive something.

The reason why we can make do with only three or four objects extracted from the blooming buzzing confusion of the world is that these few objects are made up of exactly what we need to help us perform the task of the moment. Each is a temporary nexus of meaning and action. Sometimes nexus objects are held in mind for a second or two; sometimes they only last for a tenth of a second. Th e greatly limited capacity of visual working memory is a major bottleneck in cognition, and it is the reason why we must often rely on external visual aids in the process of visual thinking.

Response patterns are the essence of the skills that bind perception to action. But they have their egative side, too. Th ey also cause us to ignore the great majority of the information that is available in the world so that we often miss things that are important.

We use the word attention to describe top-down processes. Top-down processes are driven by the need to accomplish some goal. Th is might be an action, such as reaching out and grasping a teacup or exiting a room.

If we are looking for red spots then the red spot detectors will signal louder. If we are looking for slanted lines then slanted line feature detectors will have their signal enhanced. This biasing in favor of what we are seeking or anticipating occurs at every stage of processing. What we end up actually perceiving is the result of information about the world strongly biased according to what we are attempting to accomplish.

The goal of information design must be to design displays so that visual queries are processed both rapidly and correctly for every important cognitive task the display is intended to support .

Repeating a tracing operation will take less cognitive effort, and require fewer fixations, than finding it in the first place. A hallmark of visual thinking is that it is often easier to redo some cognitive operation than to remember it.

When the eye arrives at a point of fixation, a process of visual testing begins, and patterns within the central region of the visual field are evaluated at a rate of about twenty per second; although since the eye only stays in one place for less than two-tenths of a second, roughly one to four simple patterns may be evaluated on each fixation.

The are visual aids for our brains which don't remember well. It's "things" from the external world that make us smart and visualizations serve as cues to activate this.

Because about half of our brains are wired for perception (seeing) and visualizations take advantage of that. Visualizations allow elements to pop out at us vs. text which gives us something to think about (important point).

Although it does have nice graphic design, it generally fails as an effective data visualization because: (1) It doesn't quickly communicate; (2) it forces the user to use cognitive rather than visual; (3) the 35% is incorrectly scaled and distorts the data. Good data visualizations have more to do with data integrity than "cool" design from the graphics world realm.

“Well-designed presentations of interesting data are a matter of substance, of statistics, and of design.”

(1) The scales are distorted; (2) The design "invites" you to almost compare them equally; (3) the superimposed picture doesn't add anything to making the data understandable--it's almost a gimmick; (4) scales for bar graphs should always start at zero: If -$11,000 is where it's shown, then the bottom of the graph in this perspective is actually at -$4,200,000.

It's distorting the data with graphics that aren't accurate or representative of scale

The barrels are not to scale based on the actual data. This is a problem because our visual system is dominant and automatically distorts the actual numerical scale / ratio of increase in price. In this case, the 2D barrels are processed as "3D" volumetric measurements that are grossly distorted.

1. Clear, detailed, and thorough labeling should be used to defeat graphical distortion and ambiguity.

2. The representation of numbers, as physically measured on the surface of the graphic itself, should be directly proportional to the numerical quantities represented.

3. Most important: Show data variation, not design variation.

(1) The pie chart distorts the data in Jobs' favor; (2) The pie chart makes it difficult to VISUALLY compare--in an instant--the relative rankings of Smartphone market share; (3) A simple bar graph is the solution because it uses the visual system to quickly process the relative rankings.

Pie charts measure the composition and components of set of data--and are "flashier"; bar charts rank the components in a clear, relative fashion and often communicate with more integrity. Because of this, bar charts are easier to process visually. The goal is to take advantage of the visual system so that the data pops out.

We want to maximize the ratio of data to ink being used. This graphic is still difficult to interpret. There is a lot of wasted ink in using 3D and creating drop shadows. Is it more compelling than than a 2D bar chart? Yes. Is it more communicative? No. The rule is: Use as little ink as possible to show the data.

Remove background, horizontal lines, box around graph, even horizontal line at zero. This improved version even uses negative space (white lines intersecting bars) to show the y-axis delineations.

There is value in deploying chartjunk in certain circumstances because, as research shows, chartjunk--appropriately used--can be more attractive, entertaining, and/or memorable. But at the sake of data integrity over entertainment, chartjunk should never be used.

Tufte's 'increase data density' means to put as much visual information into a chart for the given area as you can. Professor Pfister disagrees somewhat (so do I). I believe you want to put as much as necessary to communicate your point(s) (don't overwhelm). This is very germane to mobile.

Spark lines were invented by Edward Tufte. They're tiny line charts (often with values nearby) that can be fit right into the text. They can depict an overall trend, such as a trend, in a very small space.

• Maximize data-ink ratio
• Avoid chart junk
• Increase data density
• Layer information (with color, popups, etc.)

n.b. Tufte would hate this graphic -- get rid of that background because it doesn't add anything...

Not necessarily. It depends on audience, context, and goals. These examples will cause the reader to pause because it might attract some readers. But for a government audience, for example, these are out of place.

(1) Present all the data that is needed for the audience to see and understand what’s meaningful; (2) present nothing that isn’t needed; (3) rrepresent data accurately; (4) represent data in a way that is easy for the eyes to perceive and the brain to interpret; (5) provide appropriate context for interpreting the meaning of the data.

Professor Pfister believes Tufte is fundamentally right about effectiveness, integrity, and truthfulness in data viz.

Acronym = CRAP (C.R.A.P.):

1. Contrast--if two things are different, make them REALLY different (don't be a wimp)

2. Repetition--Repeat some aspects of the design throughout the entire piece (e.g., where/how bullets, headings, bold, italics, indents, etc. are used)

3. Alignment--Nothing should be placed on the page arbitrarily. Every item should have a visual connection with something else

4. Proximity--Group related items together... as
physical closeness implies a relationship.

1. Data/ink ratio is low: too much saturated and bright (loud) distracting color and unnecessary graphics; contrast is very high--yet two of the purples are similar

2. Chart junk: the spiral takes some time to figure out, even though it does draw the eye into the design; it probably distorts the data, as well.

3. Data density is on the low side

4. Little information is layered here other than the high contrast with colors

5. The fonts seem mixed and not a good choice.

The graphic above is an improvement based on Tufte's design principles.

"Clutter is a failure of design, not an attribute of inspiration."

(1) Hypothesis generation; (2) hypothesis confirmation; (3) presentation / communication

(1) Computational capacity; (2) human perceptual and cognitive capacity (these are finite human resources); (3) display capacity

We store surprisingly little information internally in visual working memory, leaving us vulnerable to change blindness: the phenomenon where even very large changes are not noticed if we are attending to something else in our view.

(1) Analyze the need; (2) analyze current solution; (3) watch users using current solution; (4) identifying problems of current model; (5) trial and error; (6) teamwork

Empathize, define, ideate, prototype, test

Observe and engage users to understand their situations

Synthesize empathy findings into insights, needs, and challenges

Brainstorm ideas for the defined problems and empathy challenges

Select ideas and create rough mockups to test viability

Get feedback on prototype to refine and improve ideas

Step 1 is the "target": What is the problem to work on? Lean about users' goals and kinds of data--then get and clean data. Then you need to "translate" that data to structure and characterize the data from which you create an abstraction of the problem--and from there, transform the data computationally. Only once you've done this can you go into the design phase. Then comes implementation. Lastly is validation.

It's the "sticky note" concept to group data... More formally, it's a group decision-making technique designed to sort a large number of ideas, process variables, concepts, and opinions into naturally related groups. These groups are connected by a simple concept. These ideas can be posted on Post-it note pads for eventual grouping, sorting, pattern finding, etc.

(1) Our eyes are task directed; (2) cameras have evenly distributed pixels whereas our eyes pixel focus is concentrated; (3) we have a brain which does visual processing in a hierarchical manner ; (4) in the brain, we get what we focus on, e.g., finding a red dot; (5) the brain is both bottom-up and top-down.

In bottom-up, the brain processes what we see up into the brain into cognition. We see low-level features which get processed in the brain as, "That's a dog..."

In top-down, the process starts from cognition in the brain down in HOW we see things. Optical illusions are top-down: The brain is wired a certain way and creates an illusion that something is appearing a certain way--when it is not.

Both happens simultaneously. In bottom-up, information drives pattern building: An image comes in through the eye then patterns or shapes emerge then they are converted to "objects" in our mind from which we form ideas, opinions, actions, etc. Top-down attentional process reinforce relevant information, i.e., in a very simple sense, we get what we are looking for. So when SEEING a dog, that information gets processed bottom up--and when we look back at the dog, the information about the dog becomes clearer (top-down).

The brain COMBINES both bottom-up and top-down information. Therefore, we have to pay attention to C.R.A.P. principles (contrast, repetition, alignment, and proximity) but we also have to keep in mind 'what will the user pay attention to'?

People are designing visualizations because they can (or have to), not because they're good at it.

Innovative, makes a product useful, aesthetic, makes a product understandable, unobtrusive, honest, long-lasting, thorough, environmentally friendly, as little design is possible

If you have continuous data, use a continuous charting format, such as a line chart. If you have discrete data, use a discrete data display format, such as a bar graph.

Bumps charts are great for comparing multiple data sources between two discrete points, e.g., comparing allocation of various channels of media spending vs. their comparative effect on purchases (see above)

Studies of shown that 3-D is a no-no it is difficult to use and interpret, especially compared to 2-D. Just don't use 3-D (except if you have a spatial relationship to show such as air flow over an aircraft wing...)

Line charts that are filled in are easier to compare than just simple line charts because cognitively, we are comparing "shapes" rather than ratios, trend, slope, etc. (see above for evolutionary comparison, left to right, of an actual design process...)

It's roughly 80% / 20%: More time must be spent on characterizing and abstracting the problem vs. visualizing a solution and design.

Its' iterative (see above)

The waterfall model is a popular version of the systems development life cycle model for software engineering. Often considered the classic approach to the systems development life cycle, the waterfall model describes a development method that is linear and sequential. Waterfall development has distinct goals for each phase of development.

Disadvantage: It constrains users from going back to prior steps in the process. Designers are moving away from it into agile development and other methods.

Agile software development is a group of software development methods based on iterative and incremental development, where requirements and solutions evolve through collaboration between self-organizing, cross-functional teams. It promotes adaptive planning, evolutionary development and delivery, a time-boxed iterative approach, and encourages rapid and flexible response to change. It is a conceptual framework that promotes foreseen interactions throughout the development cycle.

It is crowdsourcing service (hosted by Amazon) that pays people to do various computer tasks. For example designers can visually test their models by paying small amounts to crowdsourced users to perform evaluations. On the receiving end, it lets people earn money for doing small computer tasks like eyeballing and commenting on photos, answering questions about websites, etc.

Text, time series, tabular data (spreadsheets / database), images, sound, networks (relationship diagrams), fields (vectors in multiple dimensions), maps

Data model describes the data (integers, floats, operators +, -, *, etc.) Conceptual models relate to semantics. For example, data model of a one-dimensional measurement of data coming in from a sensor over time is, in the conceptual realm, temperature measurement.

Measurement, in the broadest sense, is defined as the assignment of numerals to objects or events according to rules.

In diagram, "Q" = quantitative. Regarding Q-interval, we don't care where the zero is, for example, as in dates. What we care about is the difference between dates. For Q-Ratio, we do care where the zero is; this is typically for scientific data (e.g, length, weight, etc.)

For nominal, we only care if something is equal or not. For ordinal, this is where we compare-- is something larger or smaller? In interval we add addition or subtraction so we can, for example, compute distances. In ratio, we add multiplication and division to calculate proportion.

(1) We consider the data coming in; (2) from that we derive a conceptual model, for example, temperature; (3) and from that we decide on the data type. For example, if measuring temperature do we need to round to four figures (quantitative); or do we consider the temperature is either hot, warm, or cold (ordered); or do we take a more binary approach and say that the toast is either burned or not burned (nominal).

Brightness and saturation. Hue is best for nominal data--so you can use hue to label data.

Bar charts (violates Tufte's principle of data-to-ink ratio); maps -- but not to be confused with maps that show contours or use contours to show elevations as part of a shape, like of a rock.

Properties of the image should match the properties of the data (Tufte)

Encode the most important information in the most effective way

Is the mechanism in our brains that when we go for example searching for tomatoes, it's as if the brain is saying, "All of you read sensitive cells you all have permission to shout louder. All you blue– and green–sensitive cells, try to be quiet." The same biased shouting mechanism also applies to any of the feature types processed by the primary visual cortex, including orientation, size, and motion.

Th e simple features that lead to pop out are color, orientation, size, motion, and stereoscopic depth.

Something that pops out can be seen in a single eye fixation and experiments show that processing to separate a pop-out object from its surroundings actually takes less than a tenth of a second. Things that do not pop out require several eye movements to find, with eye movements taking place at a rate of roughly three per second. Between one and a few seconds may be needed for a search. These may seem like small differences, but they represent the difference between visually efficient at-a-glance processing and cognitively effortful search.

Trying to find the target a son to features is called a visual conjunctive search, and most visual conjunctions are hard to see. The green squares do not show a pop-out effect, even though you know what to look for. The problem is that your primary visual cortex can either be tuned for the square shapes, or the green things, but not both.

Features that do pop out are hardwired into the brain. As it relates to trading, for example, with practice experts can interpret patterns that non-experts fail to see. But this expertise applies more to identifying patterns once they have been fixated with the eyes, and not to finding those patterns out of the corner of the eye .

A design to support a rapid visual query for two different kinds of symbols from among many others will be most effective if each kind of query uses a different channel. We can use shape coding for one and color coding for the other, for example. Also, we can hierarchically subordinate one element to another by adding, let's say, three channels to one element and to channels to the other, thereby making them both pop but one more than the other.

It likely comes from our need for safety when our surroundings change, for example, movement in the brush in the Savannah will help us from becoming some creature's lunch. Our sensitivity to static detail. Very rapidly away from the central fovea. Our sensitivity to motion. Much less, so we can still see something is moving out of the corner of our eye, even though the shape is invisible.

By converting the text to a uniform decision making structure using visual queues. Your Decision Bars was a step in this direction. All indicators could conceivably be pumped into the decision bar format which could either be displayed at the bottom of the chart on a bar-by-bar basis--or in a right hand margin of symbols from which the trader can act on from a top-down cognitive basis.

By first removing all prior signals unless user is in a research mode. Secondly, by placing signals in a predictable area so trader is not constantly scanning the chart landscape for a signal. (See Visual Thinking for Design, pp. 43-44.)

Finding the boundaries of an object is an important function of the pattern processing systems. In order for the brain to find an object, it must somehow be distinguished from other objects in the environment, and often the most important piece of information is that it has a continuous contour running all around it. In the trading realm, that means the current bar you are looking at is almost indistinguishable from others nearby and it takes comparatively more cognitive energy to resolve what you are seeing.

The process of combining different features that will come to the identified as parts of the same contour or region is called binding. There is no such thing as an object embedded in an image; purchase patterns of light, shape, color, and motion. Objects and patterns must be discovered, and binding is essential because it is what makes disconnected pieces of information into connected pieces of information.

As a general rule, like interferes with like. This is easy to illustrate with text as shown above. To minimize this kind of visual interference (it cannot be entirely eliminated), one must maximize feature level differences between patterns of information.

When there are multiple layers of information we can attentionally choose to focus on just one and the others will fade into the background.

Objects in the real world have structure at many scales. In the garden, for example, individual flowers provide small-scale patterns, and these are organized into patches of color depending on the design of a flower bed. The entire structure of lawns, flowerbeds, trees, and pass form a large-scale pattern.

Relationships between meaningful graphical entities can be established by any of the basic pattern-defining mechanisms: connecting contours, proximity, alignment, and closing contour, color, texture, and common movement.