Course Overview
Visualization principles across three platforms — from declarative to programmatic
Scientific Data Visualization · Summer 2025 · UNT · Prof. Dr. Zeenat Tariq
Shipping cost visualisation across Furniture, Office Supplies, and Technology sub-categories in Tableau. Key finding: Copiers have the highest shipping costs in Technology at £159,496 — visible only when broken down by sub-category, not category. This experience of visualisation as a discovery tool (revealing what summary statistics hide) applies directly to medical imaging cohort analysis, where aggregate performance metrics routinely obscure subgroup patterns.
D3.js via Observable: Iris dataset (150 samples, 3 species). Pie chart of perfectly balanced class distribution; multivariate analysis of sepal/petal dimensions using D3 interactive elements. Observable notebook: observablehq.com/d/aa9eb9017a90976c
Power BI: Tech layoffs dataset. Stacked column chart (companies × total layoffs, descending). Pie chart of layoffs by company filtered to Africa, Australia, South America — regional contribution with percentage labels.
Key learning — D3.js makes the structure of every chart visible
Tableau and Power BI abstract the chart into drag-and-drop operations. D3.js requires explicitly specifying how data values map to pixel positions, colours, and shapes — scales, axes, selections, and data binding. This gave a fundamentally different mental model: a visualisation is a function from data to visual marks. Directly applicable to designing custom clinical data visualisations that standard tools cannot produce — e.g., patient trajectory overlays, dose-response curves, or multi-site calibration plots.