Data Visualization (AD-703 (A)) Unit 2 Notes

How unit 2 is examined

This unit covers the families of visualization techniques (pixel, geometric, icon, hierarchical, scalar, contour, vector) and exploratory data analysis; the marks sit in pixel-oriented techniques, visualizing complex data, and EDA.

Pixel-Oriented Visualization Techniques

Low weight

Definition. Pixel-oriented visualization maps each data value to one coloured pixel and arranges the pixels on the screen in a chosen order, so that a very large dataset can be seen at once.

Key points.

Each attribute gets its own small window, and every record in it is one pixel, so a screen of a million pixels shows a million values.
The value is mapped to colour: for example light colour for low values and dark colour for high values, which lets patterns and outliers stand out.
The arrangement order matters: recursive pattern, circle segments and spiral (space-filling curve) layouts keep related records close together.
Circle segments divide a circle into one segment per attribute, so all attributes of a record lie along the same radius.
Benefits: it shows the maximum amount of data with no aggregation or loss, works for large datasets, and reveals clusters, trends and correlations between attributes.

Asked: [7 marks] (Dec 2024) Briefly explain Pixel-Oriented Visualization Techniques and discuss its benefits.

Geometric Projection Visualization Techniques

Definition. Geometric projection techniques project multidimensional data onto a 2-D display using geometric transformations, to find interesting views of the data.

Key points.

A scatter plot shows two attributes on the x and y axes, and a scatter plot matrix shows every pair of attributes.
Parallel coordinates draw one vertical axis per attribute and each record as a polyline crossing the axes.
Projection pursuit and prosection views help to find useful low-dimensional projections.
Very many dimensions or records cause clutter and overlapping lines.

Icon-Based Visualization Techniques

Definition. Icon-based techniques represent each record by an icon (glyph) whose visual features, such as shape, size, colour or angle, encode the attribute values.

Key points.

Chernoff faces map attributes to facial features such as eye size, mouth curve and nose length, using our skill at reading faces.
Stick figures use limb angles and length for attributes, and two attributes fix the position on the plane.
Shape coding and colour icons are other variants.
It suits small to medium data with few attributes, and needs learning to read.

Hierarchical Visualization Techniques

Definition. Hierarchical techniques split the dimensions into subsets and show them in nested or tree-structured views, so that hierarchy or part-whole structure is visible.

Key points.

A treemap fills a rectangle with nested rectangles whose area shows the value of each node.
Worlds-within-worlds nest one plot inside another for the remaining dimensions, and dimensional stacking embeds one axis grid inside another.
A sunburst is the radial version, with rings for levels of the hierarchy.
It suits organisational, file-system and category data.

Visualizing Complex Data and Relations

Definition. Complex data is high-dimensional, large or networked data, and it is visualized by techniques that reduce or arrange it so that its structure and relations become visible.

Key points.

Dimensionality reduction (PCA, t-SNE) projects many attributes to 2-D; for example iris data with 4 attributes becomes a 2-D scatter showing three clusters.
Parallel coordinates show each record as a line across attribute axes, for example to compare cars on price, mileage and power.
Treemaps show hierarchical data as nested areas, for example disk usage by folder.
Network (node-link) graphs show relations, for example friendships in a social network.
Use dimensionality reduction for clusters, parallel coordinates for comparing attributes, treemaps for part-whole, and graphs for relations.

Asked: [7 marks] (Dec 2024) Explain different techniques for Visualizing Complex Data with help of example.

Scalar and point techniques, Color maps

Definition. Scalar data has one value per point (temperature, pressure); a colour map converts each scalar value to a colour.

Key points.

A rainbow colour map runs from blue (low) to red (high), while a grey scale uses brightness only.
A diverging map uses two hues around a neutral centre, which suits values above and below zero.
Point techniques plot a glyph or marker at each location, with size or colour showing the value.
A poor colour map can create false features, so the range and legend must be clear.

Contouring Height Plots

Definition. A contour (isoline) is a curve joining points that have the same scalar value; a height plot shows the value as height above a plane.

Key points.

Contour maps show terrain elevation through lines of equal height, as in topographic maps.
Closely spaced lines mean steep change, and widely spaced lines mean gentle change.
Filled contours colour the bands between isolines, and 3-D height plots draw the surface as a mesh.
Isolines are computed by interpolating the grid values at the chosen level.

Vector visualization techniques, Vector properties, Vector Glyphs

Definition. Vector data assigns a magnitude and direction to each point, as in wind or fluid flow, and vector glyphs draw this as arrows or similar shapes.

Key points.

Vector properties are magnitude, direction, divergence (source or sink) and curl (rotation).
A hedgehog plot draws an arrow at each grid point, with arrow length showing magnitude and orientation showing direction.
Other glyphs are cones, needles and warped arrows.
Too many arrows clutter the display, so sampling or scaling is needed.

Vector Color Coding Stream Objects

Definition. Stream objects are curves that follow the flow, and colour coding shows the magnitude or another scalar along them.

Key points.

A streamline is tangent to the velocity field at every point at one instant.
Stream ribbons, stream tubes and stream surfaces show twist, spread and expansion of the flow.
Colour is mapped to speed or pressure, for example red for fast and blue for slow.
Seed points decide where the streamlines start.

Exploratory data analysis (EDA) Techniques

Definition. Exploratory data analysis is the initial investigation of a dataset, using summary statistics and graphs, to find patterns, outliers and relations and to form hypotheses before modelling.

Key points.

Summary statistics (mean, median, standard deviation) describe the centre and spread, for example marks of a class.
A histogram shows the distribution shape, and a box plot shows median, quartiles and outliers.
A scatter plot shows the relation between two variables, and the correlation coefficient measures its strength from -1 to +1.
EDA also finds missing values and errors, and generates hypotheses that modelling then tests.

Asked: [7 marks] (Dec 2024) Discuss Exploratory data analysis techniques with the help of suitable examples.

Last-minute revision

Pixel-oriented: one value equals one coloured pixel; layouts are recursive pattern, circle segments and spiral.
Geometric projection: scatter plots and parallel coordinates.
Icon-based: Chernoff faces and stick figures encode attributes as features.
Hierarchical: treemap uses area for value; sunburst is radial.
Complex data techniques: dimensionality reduction, parallel coordinates, treemaps, network graphs.
Colour map converts a scalar value to a colour; a contour joins equal values.
Vector properties: magnitude, direction, divergence, curl.
Streamline is tangent to the flow; colour shows speed.
EDA: summary statistics, histogram, box plot, scatter plot, correlation, then hypotheses.
Correlation ranges from -1 to +1.

Memory hooks

Pixel = one value, one dot.
Chernoff = faces tell the data.
Treemap = area is size.
Contour = same height, same line.
EDA = look before you model.

Coverage checklist

Pixel-Oriented Visualization Techniques: Q Dec 2024 (7 marks).
Geometric Projection Visualization Techniques: no past question.
Icon-Based Visualization Techniques: no past question.
Hierarchical Visualization Techniques: no past question.
Visualizing Complex Data and Relations: Q Dec 2024 (7 marks).
Scalar and point techniques, Color maps: no past question.
Contouring Height Plots: no past question.
Vector visualization techniques, Vector properties, Vector Glyphs: no past question.
Vector Color Coding Stream Objects: no past question.
Exploratory data analysis (EDA) Techniques: Q Dec 2024 (7 marks).

Account

Data Visualization (AD-703 (A)) - Unit 2 Short Notes

How unit 2 is examined

Pixel-Oriented Visualization Techniques

Geometric Projection Visualization Techniques

Icon-Based Visualization Techniques

Hierarchical Visualization Techniques

Visualizing Complex Data and Relations

Scalar and point techniques, Color maps

Contouring Height Plots

Vector visualization techniques, Vector properties, Vector Glyphs

Vector Color Coding Stream Objects

Exploratory data analysis (EDA) Techniques

Last-minute revision

Memory hooks

Coverage checklist