Charles Karpati | Data Visualization

Univariate

Core tools for distribution shape, spread, and outliers.

• Histogram – bin-based distribution profile.
• Kernel density estimate – smoothed distribution.
• Rug plot – point-level distribution cue.
• Empirical CDF / CCDF – cumulative distribution structure.
• Box plot – robust summary + outlier detection.
• Violin plot – box–density hybrid.
• Stem-and-leaf display – quick numeric snapshot.
• Bean plot – density + individual observations.
• Quantile plots (Q–Q, P–P, worm plot) – distributional deviation diagnostics.
• Ridgeline plot – stacked density comparison across groups.

Bivariate

Characterize relationships, patterns, and dependence.

• Scatter plot – baseline relational structure.
• Hexbin plot – high-N scatter aggregation.
• 2D KDE / contour plot – continuous joint density.
• Joint histogram – binned joint distribution.
• Two-sample Q–Q plot – direct distribution comparison.
• Correlogram – correlation mapping.
• Scatter with marginal densities – joint + marginal structure.
• Alpha-blended scatter – density-enhanced point cloud.

Multivariate

Expose patterns in higher-dimensional spaces.

• Scatterplot matrix (SPLOM) – all-pairs relationships.
• Parallel coordinates – multi-feature path structure.
• Andrews curves – function-based multivariate encoding.
• Star coordinates – radial multi-axis encoding.
• Glyph encodings (Chernoff faces, radial glyphs) – shape-encoded multivariate values.
• Biplot – PCA scores and loadings.
• t-SNE / UMAP embeddings – nonlinear manifold projections.
• Radar chart – profile comparison across features.
• Heatmap – matrix visualization for features × samples or correlations.
• Mosaic plot – multiway categorical dependence.

Categorical

Compare counts, proportions, and structure.

• Bar chart – category comparisons.
• Stacked bar chart – part–whole segmentation.
• Grouped bar chart – multi-factor comparison.
• Dot plot (Cleveland) – high-resolution categorical comparison.
• Lollipop chart – stem–value alternative to bars.
• Pie chart – coarse proportion cue.
• Marimekko (Mekko) chart – joint categorical proportions.
• Contingency table – joint distribution summary.
• Alluvial plot – category transitions.
• Sankey diagram – categorical flow structure.

Time Series

Reveal temporal evolution, stability, and autocorrelation.

• Line plot – baseline continuous-time trajectory.
• Run chart – level shifts over time.
• Control chart – stability and process limits.
• Seasonal subseries plot – seasonal pattern decomposition.
• Lag plot – serial dependence diagnostics.
• ACF/PACF – autocorrelation structure.
• Spectral density plot – frequency-domain behavior.
• Horizon chart – compressed multi-band amplitude view.
• Fan chart – forecast paths with uncertainty bands.
• Calendar heatmap – long-horizon temporal distribution.
• Rolling-window metrics – local variance/mean/volatility.

Model and Residual Diagnostics

Assess fit quality, calibration, and assumptions.

• Residuals vs fitted – bias or heteroscedasticity.
• Residual Q–Q – normality deviations.
• Scale–location plot – variance structure.
• Influence/leverage plots – Cook’s distance and hat values.
• Partial dependence plots – marginal effect visualization.
• ICE curves – individual conditional effects.
• Calibration plot – probabilistic reliability.
• ROC / PR curves – classification discrimination.
• Lift chart – ranked predictive performance.
• Variable-importance heatmap – feature contribution profiling.
• Forest plot – effect sizes with intervals.

Network and Non-Hierarchical Structures

For relational data lacking parent–child nesting.

• Force-directed graph – topology-driven layouts.
• Adjacency matrix heatmap – matrix-encoded network structure.
• Edge-bundled network – reduced visual clutter via bundling.
• Chord diagram – directional/bilateral flows.
• Sankey diagram – flow-oriented network representation.

Hierarchical Structures

For nested, tree-based, or clustered data.

• Tree diagram – explicit hierarchy.
• Cluster dendrogram – agglomerative/divisive clustering.
• Treemap – space-filling hierarchy.
• Sunburst – radial hierarchical partition.
• Icicle plot – stacked hierarchical partition.
• Partition diagram – recursive splitting structure.
• Packed circles – containment hierarchy.

Text Visualization

Frequency- and structure-oriented tools for tokenized text.

• Word cloud – size-encoded term frequency; coarse quantitative precision.
• Term–document heatmap – frequency matrix structure.
• Co-occurrence network – relational term structure.
• Topic-embedding scatter (t-SNE/UMAP) – semantic-space projection of documents/topics.
• Concordance / KWIC strips – positional term context.
• N-gram bar charts – frequency of multi-token sequences.

Spatial

Geospatial distribution, density, and structure.

• Choropleth – region-level value encoding.
• Proportional symbol map – magnitude-encoded markers.
• Dot-density map – discrete-event spatial distribution.
• Hex-map – normalized spatial binning.
• Kernel density surface / heatmap – smoothed spatial intensity.
• Cartogram – value-distorted geography.
• Flow map – directional movement.
• Voronoi tessellation – spatial partition via nearest regions.
• Contour/elevation map – gradient-based spatial fields.

A box plot encodes the following elements:

• Range – the span from the minimum to the maximum observed values.
• Interquartile range (IQR) – the central 50% of the data, defined as IQR = Q3 - Q1.
  • Q1: 25th percentile, median of the lower half.
  • Q3: 75th percentile, median of the upper half.

Outliers follow the standard 1.5 * IQR rule: any observation below Q1 - 1.5 IQR or above Q3 + 1.5, IQR is classified as an outlier.

Source: https://www.amazon.com/Visual-Display-Quantitative-Information/dp/1930824130

there are six principles to ensure Graphical Integrity:

• Make the representation of numbers proportional to quantities
• Use clear, detailed, and thorough labeling
• Show data variation, not design variation
• Use standardized units, not nominal values
• Depict ’n’ data dimensions with less than or equal to ’n’ variable dimensions
• Quote data in full context

Source: https://realpython.com/python-data-visualization-bokeh

Histograms and Density Plots

Histograms work very well for display a single variable from one category.

For displaying multiple categories use

• Side-by-Side Histograms
• Stacked Histograms
• Density Plots
• Rug Plot