LoginSign Up
← Back to all sparks
F

factoextra

ANALYTICS
Velocity3.8

Extract and Visualize the Results of Multivariate Data Analyses

github.com/kassambara/factoextra ↗

factoextra woke from six years of silence and stopped being a FactoMineR front-end

dimension reductionrggplot2clusteringtidymodelsvisualization
◆Current state
factoextra supplies the fviz_* grammar most R users reach for when plotting PCA, correspondence analysis and clustering results. It sat effectively dormant from 2020 to early 2026, then shipped three releases in five months that resolved a six-year issue backlog, decoupled it from its original backends, and taught it to plot UMAP and t-SNE embeddings. The current version also reaches into tidymodels, plotting a PCA fitted inside a recipe or workflow directly.
◆Where it's heading
The direction is from wrapper to visualization layer. as_factoextra_pca() turned the fviz_pca_* family into something that consumes coordinates from anywhere — cmdscale, vegan, a custom decomposition, a recipes step — rather than only FactoMineR objects, and the vignette on extending factoextra to new backends says the maintainer intends others to plug in. The embedding support shows the same care: no scree plot, no correlation circle, and a convex hull instead of a confidence ellipse, because an embedding does not preserve the metric an ellipse would assume.
◆Prediction
With recipes and workflows already wired in, the next likely step is covering more tidymodels dimension-reduction steps (step_umap, step_ica, step_pls) through the same as_factoextra_pca() entry point rather than adding new fviz_* functions per method.

◆Recent moves

  1. 22d ago

    factoextra 2.2.0 adds UMAP and t-SNE plotting, plus tidymodels recipes

    ⚡ SPARK

    The release where factoextra stops being a linear-factor tool. fviz_umap() and fviz_tsne() bring nonlinear embeddings into the same grouping-and-ellipse grammar, and the extension point opened in 2.1.0 grows recipe and workflow methods.

    View source ↗
  2. 1mo ago

    factoextra 2.1.0 decouples its plots from any specific backend

    ⚡ SPARK

    The hinge of the revival: as_factoextra_pca() builds an fviz-ready object from pre-computed coordinates, so the plotting family no longer requires a FactoMineR or prcomp object behind it. Everything 2.2.0 does with embeddings and tidymodels rides on this.

    View source ↗
  3. 5mo ago

    factoextra 2.0.0 ends a six-year dormancy with breaking modernization

    Debt payoff rather than new direction: 30+ open issues closed, aes_string() replaced with the .data pronoun, tidyr and reshape2 dropped, and hard floors set at R 4.1.0, ggplot2 3.5.2 and FactoMineR 2.13. Worth reading the Hopkins note — get_clust_tendency() now uses the corrected Wright (2022) formula, so cluster-tendency values will not match anything computed with an earlier version.

    View source ↗
  4. 6y ago

    factoextra 1.0.7

    A single stringsAsFactors argument added ahead of R 4.0.0. The last release before the package went quiet for six years.

    View source ↗
  5. 6y ago

    factoextra 1.0.6

    Late-era maintenance from the original run: argument validation in fviz_nbclust(), corrected cluster ordering in fviz_mclust_bic(), outlier styling controls in fviz_cluster(), and a first correction to the Hopkins statistic that 2.0.0 would revisit.

    View source ↗
  6. 8y ago

    factoextra 1.0.5

    Coverage-widening from the package's first life: fviz_dend() learns diana objects, fviz_cluster() learns HCPC results, and fill.ind/fill.var plus geom controls land. Correlation circles also stop rendering as ellipses — the kind of geometric honesty the 2026 embedding work carries forward.

    View source ↗