quollr: Visualising How Nonlinear Dimension Reduction Warps Your Data
Source:R/quollr-package.R
quollr.Rdquollr helps you assess nonlinear dimension reduction (NLDR) layouts by constructing a 2-D hexagonal-bin wireframe model and lifting it into high-dimensional space for visual inspection. This makes it possible to determine which NLDR method and hyper-parameter choices produce the most faithful representation of high-dimensional data.
Details
The main workflow is:
Fit the model with
fit_highd_modelCombine data and model with
comb_data_modelInspect in high dimensions with
show_langevitour
Author
Maintainer: Jayani P. Gamage jayanilakshika76@gmail.com (ORCID)
Authors:
Dianne Cook dicook@monash.edu (ORCID)
Paul Harrison paul.harrison@monash.edu (ORCID)
Michael Lydeamore michael.lydeamore@monash.edu (ORCID)
Thiyanga S. Talagala ttalagala@sjp.ac.lk (ORCID)
Examples
# Fit the complete model pipeline
model <- fit_highd_model(
highd_data = scurve,
nldr_data = scurve_umap,
b1 = 15,
q = 0.1,
hd_thresh = 0
)
#> ✔ Model generated successfully!!!
# Combine data and model for visualisation
df <- comb_data_model(
highd_data = scurve,
model_highd = model$model_highd,
model_2d = model$model_2d
)
# Evaluate goodness of fit
glance(x = model, highd_data = scurve)
#> # A tibble: 1 × 2
#> Error HBE
#> <dbl> <dbl>
#> 1 265. 0.157
# Inspect the lifted model in high dimensions (interactive only)
if (FALSE) { # \dontrun{
show_langevitour(
point_data = df,
edge_data = model$trimesh_data
)
} # }