Quality Assessment for Interval-Based Dimensionality Reduction
QAIDR provides tools for evaluating how well interval-based dimensionality reduction (DR) methods preserve the structure of high-dimensional interval data, using a co-ranking matrix framework.
Features
- 4 interval distance metrics: Interval Euclidean, Hausdorff, Ichino-Yaguchi, L2-Wasserstein
- 6 co-ranking indices: Quality (Q) and Behavior (B) variants of Trustworthiness & Continuity, MRRE, and LCMC
- 6 DR method wrappers: C-PCA, V-PCA, MR-PCA, SPCA, IMDS, Int-UMAP
- Permutation tests for statistical significance
- Visualization: 2D projection plots and K-neighbourhood profile plots
Installation
Install the development version from GitHub:
# install.packages("pak")
pak::pak("hanmingwu1103/QAIDR")Quick Start
library(QAIDR)
# Load and standardize the built-in Cars dataset
data(cars_mm)
x <- standardize(cars_mm)
# Compute interval distances
D <- idist(x, metric = "Wasserstein")
# Run all 6 DR methods (requires symbolicDA, RSDA)
proj <- run_idr(x)
# Assess quality across all method-metric combinations
result <- assess_quality(x, proj, K = 5, perm_test = TRUE, n_perm = 1000)
print(result)
# Visualize
plot_projections(proj, labels = cars_mm$labels)
profiles <- k_profiles(x, proj)
plot_k_profiles(profiles, metric = "Wasserstein")Documentation
- Package website
-
vignette("introduction")– Basic workflow -
vignette("real-data-analysis")– Reproduces the real data analysis -
vignette("simulation-study")– Reproduces the simulation study