IS6335 Data Visualization

How IS6335 is assessed

The syllabus for Semester A, 2026/27 records the assessment as continuous at one hundred per cent, lists these four tasks, and prints the hundred itself. Two details are worth knowing. The syllabus names the thirty per cent component Assignments, while the teaching plan, the schedule and every deck call the same component lab exercises; they are one component, not two. And the internal splits reconcile: six exercises at five per cent is thirty, and a milestone report at two plus a milestone presentation at two plus a final presentation at sixteen is twenty. There is no examination. No hurdle, minimum component mark or separate pass condition appears anywhere in the course materials available here, which is an absence of a statement rather than a statement of absence, so confirm the current pass conditions on Canvas.

What IS6335 covers

Twelve chapters follow the published module sequence, from visual reasoning and the what why how framework through tables, view manipulation, spatial data and networks into computer vision, then the tutorial toolchain and the two projects that carry sixty per cent of the course.

The syllabus is explicit that it carries no prerequisite of programming background, and the week one materials repeat that none is assumed.

What it does assume is that you are willing to defend a design decision.

The published goal is to learn how to use visualization tools for data interpretation in a business context, moving from the graphical user interface of Tableau to programmable packages in R and Python, and then to apply those concepts and skills in a final project.

The three intended learning outcomes carry their own weights: describing and gaining insight into the theory of visual presentation and the use of visual reports in business communication and analytics at forty per cent, applying Tableau skills at twenty, and applying programmable tools and using visualization to augment machine learning and statistical analysis at forty.

Everything is continuous assessment

There is no examination.

The syllabus records the assessment as continuous at one hundred per cent, and the four tasks are class participation at ten per cent, assignments at thirty, the group project at twenty and the individual project at forty.

The practical consequence is that nearly three quarters of the course is earned on work you design yourself, and that preparation means building the ability to justify a choice rather than the ability to recall a definition.

The internal splits are published and they reconcile.

The thirty per cent assignments component is six weekly lab exercises at five per cent each, placed in weeks three to six and weeks eight and nine, with week seven given over to the group milestone presentation. The twenty per cent group component is a milestone report at two per cent, a milestone presentation at two and a final presentation at sixteen.

The forty per cent individual project is a written report, taking the form of a short research paper and marked against a five criterion rubric worth one hundred points.

The spine of the course is one framework

Week two introduces a design analysis framework that the rest of the course keeps returning to.

It splits any visualization decision into three questions: what is shown, which produces a data abstraction; why the user is looking at it, which produces a task abstraction; and how it is shown, which produces the idiom.

The vocabulary is deliberately abstract, because a description written without domain words lets a solution travel from one problem to an unrelated one.

The three modules then work through that framework. Module one builds the conceptual basics: why visualization works at all, the framework itself, and marks and channels with the colour rules that follow.

Module two takes the three dataset types the course recognises, moving through tables and their idioms, view manipulation and large data, spatial and three dimensional data, and networks with adjacency matrices, community detection and text mining.

Module three opens computer vision and deep learning, where the image is a dataset type you already know and the model's behaviour becomes something that itself has to be visualized before it can be trusted.

Where students lose marks, and where this guide concentrates

The largest rubric criterion is worth forty of the individual report's hundred points and it names four separate things: the content quality of the visual analysis, the application of visualization concepts, the steps of analysis, and the description of findings and insights.

A report full of competent charts with no account of why those idioms were chosen is answering one of those four. That is the most common and most expensive failure in a course like this one, and it is avoidable with a habit: for every figure, write one sentence naming the attribute type, the channel you gave it and the task it serves.

The second common failure is starting from a dataset rather than from a question.

Twenty points depend on the significance of the problem statement and twenty more on research and practical implications, and neither can be recovered later from data that was explored without a purpose.

The course's own tutorial document makes the point structurally: before any chart is built it asks what question you want to answer and what the goal is, and only then which data and which chart.

How the twelve chapters are organised

The chapters follow the published weekly plan.

Chapters one to four cover module one, from the definition of a visualization system and the test for when one is worth building, through the framework, the dataset and attribute typologies, and the assignment of marks and channels.

Chapters five to nine cover module two: table idioms, the four view manipulations and what to do when there are more marks than pixels, spatial data and the choropleth problem, network structures and topology, and then matrices, communities and text. Chapter ten covers the advanced module.

Chapter eleven follows the tutorial track across all three tools, including the Tableau object model and the data typing failure the course's own tutorial walks you into. Chapter twelve works through both projects against the University's published requirement lists and the rubric.

Two published figures in the course documents disagree with each other, and both are current.

The teaching plan describes the individual report as ten to fifteen pages while the week one lecture slides describe a comprehensive written report of five to ten pages. Group size is given as five to eight in the teaching plan and in the week one deck, and as six to eight in the week two deck.

Where two current documents disagree, take the figure from the brief posted for your own section on Canvas, and this guide flags the disagreement rather than choosing for you.

Worked example · free

Take one dataset through the framework, the channels and the report criteria

Key terms

Visualization system

A computer-based system that puts a dataset into visual form so that a person can do a task better than they could otherwise. The definition is built around a task, which is why effectiveness in this course is always effectiveness for something.

Data abstraction

The answer to the what question: which of the three dataset types you hold, and the type of every attribute in it. It is written without domain words so that two unrelated problems can be compared.

Task abstraction

The answer to the why question, written as an action and a target. Discovering an outlier and presenting a known trend are different tasks over the same table and need different designs.

Idiom

One distinct way of building a visual representation and of letting a reader act on it, such as a bar graph, a pie chart or a network. The word covers interaction as well as encoding, so a filterable chart is a different idiom from a static one.

Mark

The basic graphical element standing for an item: a point, a line, an area or a volume. Choosing the mark fixes which channels are then available to carry attributes.

Channel

A property of a mark that can be varied to carry a value, such as position, length, hue, saturation, luminance, size, angle, curvature or motion. Channels differ in how accurately a reader decodes them.

Attribute type

Whether an attribute is categorical, ordinal or quantitative. The type decides which channels may legitimately carry it, and a numeric identifier is categorical however it is stored.

Ordering direction

Whether an ordered attribute is sequential, diverging or cyclic. It decides the shape of a colour scale, and a diverging attribute drawn on a sequential ramp loses the only boundary that matters.

Derive

Creating a new attribute by transforming the ones you were given, rather than drawing the file as it arrived. It is named as one of the major strategies for handling complexity, and the column a question needs is often one that does not yet exist.

Adjacency matrix

A network written as a grid with every node on both axes and a filled cell wherever two nodes are connected. Its layout does not depend on the links, so it stays readable at any density.

Continuous field

A spatial dataset given as a grid of positions, each cell carrying attributes. Its positions are samples rather than choices, so resolution is a constraint you inherit rather than a design decision.

Dimensions and measures

The Tableau division between fields that cannot be aggregated, normally used for row and column headings, and fields that can be calculated or aggregated and are normally plotted. The two convert into each other, and that conversion repairs a large family of wrong charts.

IS6335 FAQ

Questions students ask in each IS6335 chapter

Why does the course spend a whole lecture on why rather than how?

Because the test for whether to visualize at all decides everything downstream, and it is the one step that cannot be recovered later. A chart built for a question that was already fully specified answers nothing a rule could not answer faster, and a chart built with no stated task cannot be evaluated at all. The two criteria introduced in the same week, information density and readability of spatial position, are also the criteria the course returns to when it compares designs, so the material is not preliminary. It is the standard the later weeks are judged against. Chapter 01 · Why Visualization Works and When It Is the Right Tool

Do I have to answer the three questions in order?

You have to write them separately, and the order what, why, how is the order the lecture uses, but the process is explicitly described as iterative and tricky rather than as a single pass. In practice the task abstraction is the one that moves: a first attempt says the reader wants to compare, and working through the idiom reveals that the real target was an outlier. What matters is that the three answers stay distinct on the page, because the stack is a dependency. If the data abstraction changes, the other two have to be revisited; if only the task changes, the data description survives. Chapter 02 · The What Why How Design Analysis Framework

How do I decide whether my data is a table or a network?

By the question, not by the file. The same file of taxi trips is a table if you are asking how fares are distributed by hour, because items are trips and fare and hour are attributes; it is a network if you are asking which locations feed which, because nodes are locations and links are trips; and it is spatial if you are asking where demand concentrates. One physical file can carry three possible data abstractions, and the task decides which one is in force. What you cannot do is leave the decision implicit, because each of the three brings a different set of idioms with it. Chapter 03 · Data Abstraction, Attribute Types and Task Targets

Is a rainbow colour scale really that bad?

For an ordered attribute, yes, and the reason is structural rather than a matter of fashion. Distinctness is not the requirement; order is. Hue is an unordered channel, so a reader cannot tell which of two hues is higher without consulting the legend every time, and a rainbow ramp also has uneven perceptual steps, so equal differences in the data look unequal on the page. It then collapses entirely in greyscale. A single-hue luminance ramp keeps the order visible, keeps the steps even and survives printing. For a genuinely categorical attribute with few levels, distinct hues are exactly right. Chapter 04 · Marks, Channels and Colour Design Rules

When is a pie chart acceptable?

When there are few parts, when they genuinely sum to one meaningful whole, and when the reader's question is about share rather than about rank or precise magnitude. It is on the published idiom list, so treating it as forbidden misreads the course. What it costs you is accuracy, because angle is decoded less well than length, and a legend, because slices are usually labelled away from the marks. Two pies placed side by side are the case to avoid outright, since comparing angles across separate circles is harder than any other comparison in the week's material. Chapter 05 · Visualizing and Arranging Table Data

Should I reduce first or reorder first?

Reorder first, then reduce, because reordering is free and reducing is not. Sorting by the value the question is about costs nothing, changes no data and frequently answers the question on its own, since the reader wanted to know which items are at the extremes. Reducing removes items from the view, and every removal is a decision that has to be justified and reported. Doing the free move first also tells you how much reduction is actually needed, which is usually less than it looked before the sort. Faceting comes in only when a second attribute is genuinely part of the question rather than merely present in the file. Chapter 06 · View Manipulation and Big Data Visualization

My map of totals looks convincing. Why should I change it?

Because two separate things are already encoded before your attribute arrives. Area on the page carries land size, and a count of almost anything grows with the population that produced it, so a big populous region will dominate a total map whatever the underlying rate. That is why a map of national renewable generation makes large countries look like leaders. Divide by the population or by the region's own total to get a rate, map that, and if the absolute totals still matter operationally, put them in a sorted bar chart beside the map where length can be compared on a common baseline. Chapter 07 · Spatial Data, Maps and Three Dimensional Views

My data is a spreadsheet of emails. Is that a network?

It is already an edge list, with sender and recipient as the two ends of each link, and two derivations are needed before anything can be drawn. Repeated emails between the same pair must be collapsed into one link carrying a count, otherwise the same link is drawn hundreds of times and the picture reports volume as clutter. And a decision has to be made about direction, because sender to recipient is directed and treating the relationship as undirected merges two different things into one. Both decisions belong in the write-up, since a reader cannot see either of them in the finished drawing. Chapter 08 · Network Data Structures and Graph Visualization

My matrix looks like random speckle. What now?

Reorder the rows and columns, either by a community detection result or by a simple clustering of the rows, so that similar nodes sit together and any block structure is brought onto the diagonal. Then stop if the speckle persists under two different orderings. That persistence is evidence the network has no block structure, which is a real finding and should be reported as one rather than hidden by hunting through orderings until a flattering one appears. The temptation to keep searching is exactly the failure the stability check exists to prevent. Chapter 09 · Adjacency Matrices, Communities and Text Mining

We are not machine learning students. How much of this do we need?

Enough to read a model's output as data rather than as a verdict, which is what the third intended learning outcome asks for when it names using visualization to augment machine learning and statistical analysis. The course introduces the architectures so that the vocabulary is available, and the transferable skill is the auditing one: knowing that an accuracy figure hides class balance, that a confusion matrix recovers it, and that the same reordering move from the network week applies. Training a model well is not what a visualization report is marked on; showing what it does and does not support is. Chapter 10 · Computer Vision and Deep Learning for Images

Should I use the drag-and-drop tool or code for the project?

Decide by repetition rather than by preference. Anything you will rebuild when the data changes is cheaper in code, because the steps are written down and rerun rather than remembered and repeated. Anything you will build once and then discuss with an audience is faster by hand, and interaction is much cheaper in the tool. Most groups end up with both, and the write-up should say which parts came from where. For a weekly exercise the answer is simpler: use the tool the tutorial uses, because the exercise is testing that week's idea rather than your tooling. Chapter 11 · Tableau, R and Python Across the Lab Exercises

How long should the individual report be?

Two current course documents give different answers and neither is stale, so take the length from the brief posted for your own section on Canvas. The teaching plan describes a final report that takes the form of a short research paper of ten to fifteen pages. The week one lecture slides describe a comprehensive written report of five to ten pages. Both belong to this offering. What does not vary is the shape: the rubric is the same either way, so the allocation between sections matters more than the total, and a shorter report that gives forty per cent of its space to the visualization analysis will outscore a longer one that gives it a quarter. Chapter 12 · The Group Project and the Individual Report

How to prepare for the assessments

Carry one dataset of your own through the whole course rather than meeting a new one each week. Pick it in the first fortnight, ideally the one your group will use, and at the end of every topic add one figure to it that the week has just made possible.

Each figure gets a one sentence justification naming the attribute type, the channel you gave it and the task it serves; that sentence is the unit the individual report is marked in, and writing forty of them across the semester is far easier than writing ten of them in the revision week. Submit every lab exercise, because the marks are for submission and six of them are the entire assignments component.

Before each deadline, run the five criterion rubric over what you have as a checklist rather than rereading the draft, and delete any figure for which you cannot write the justifying sentence.

Your AI Information Technology tutor for IS6335

Stuck on a hard IS6335 question? Sia is AskSia’s AI Information Technology tutor — ask any IS6335 Data Visualization question and get a clear, step-by-step explanation grounded in how the course is actually taught and assessed. Read this whole study guide free, then take your hardest questions to Sia.