level: research
high-dimensional categorical data is common in genetics, biomedicine, and social sciences, but visualization tools lag behind those for continuous variables. existing methods often scale poorly, rely on low-dimensional projections that lose connection to the original data, or focus on prediction over interpretation. the categorical generalized association plots (cgap) framework addresses these issues by keeping the raw data matrix visible while adding interpretable structure.
cgap uses homogeneity analysis (homals) to embed both subjects and category levels into a three-dimensional space. this embedding is then mapped to red-green-blue color coordinates, so similar patterns get similar colors. the result is a heatmap where rows and columns are ordered by similarity, making clusters and relationships easier to see. the framework includes three linked views: a homals-guided heatmap of the raw data, a subject plot, and a category plot, all coordinated for interactive exploration.
the approach works with nominal, ordinal, and binary data, and scales to large datasets by using efficient algorithms. by preserving the original data values in the heatmap, analysts can verify patterns directly rather than trusting abstract projections. this makes cgap useful for tasks like identifying patient subgroups in clinical data or finding response patterns in survey research, where understanding the data structure is as important as prediction.
why it matters: it gives data scientists a practical tool to visually explore complex categorical data without losing touch with the original values, aiding pattern discovery and interpretation.