pith. sign in

arxiv: 2605.04306 · v1 · submitted 2026-05-05 · 💻 cs.HC

dtour: a steerable tour de vis through high-dimensional data

Fritz Lekschas , Nezar Abdennur This is my paper

Pith reviewed 2026-05-08 16:55 UTC · model grok-4.3

classification 💻 cs.HC
keywords high-dimensional datadata toursprojection visualizationinteractive interfacessteerable explorationdimensionality reductionweb visualizationlarge-scale data
0
0 comments X

The pith

dtour unifies static previews, geodesic scrubbing, manual control, and grand tours into one steerable browser interface for high-dimensional data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Existing tools for touring high-dimensional data force users to choose between limited control in guided paths or unstructured wandering, making it hard to explore flexibly. dtour solves this by putting static projection previews, reversible scrubbing on continuous paths, manual manipulation, and a wandering grand tour together in one interface that users can progressively explore. This matters because it lets analysts move smoothly between expert-guided views and open discovery, which is essential for understanding complex data where single views hide important details. The tool runs in browsers, handles large data, and works with common coding tools, as shown in examples with text, images, and single-cell biology data.

Core claim

We present dtour, a tour interface that combines static projection previews, reversible scrubbing along continuous geodesic projection paths, manual projection manipulation, and a wandering grand tour, all within a single progressive exploration interface. dtour scales to millions of points via GPU-accelerated rendering, runs in any modern browser, and integrates with both Python and JavaScript ecosystems. We demonstrate dtour on text, image, and single-cell data for two usage scenarios: gradually revealing structure in high-dimensional data and validating non-linear dimensionality reduction outputs.

What carries the argument

The dtour interface, which unifies static previews, reversible geodesic scrubbing, manual controls, and grand tour wandering into one progressive system.

If this is right

  • Seamless movement between guided and free exploration of projections.
  • Scalable interactive analysis of large datasets in standard web browsers.
  • Direct support for validating results from dimensionality reduction techniques.
  • Easy integration into Python and JavaScript data analysis pipelines.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar unification of exploration modes could benefit visualization tools in other scientific domains.
  • Linking dtour with automated analysis could enable hybrid human-AI data tours.
  • The web-based implementation suggests applications in educational settings for teaching high-dimensional concepts.

Load-bearing premise

The specific combination of static previews, geodesic scrubbing, manual manipulation, and grand tour features in dtour delivers practical advantages in usability and insight over prior tools without new drawbacks in performance or interaction.

What would settle it

A comparative user study measuring task performance, such as time to identify data structures and accuracy in validating reductions, when using dtour versus existing tour interfaces.

Figures

Figures reproduced from arXiv: 2605.04306 by Fritz Lekschas, Nezar Abdennur.

Figure 1
Figure 1. Figure 1: dtour’s interface for exploring high-dimensional data along a tour of keyframe projections. dtour unifies three modes of increasing projection traversal steerability. (1) dtour shows a central 2D scatter with a gallery of projection previews to give the user an overview. (2) The user can advance the central scatter along a cyclical guided tour by clicking a preview, scrubbing the slider, or scrolling to sm… view at source ↗
Figure 2
Figure 2. Figure 2: Usage Scenarios. Left: Attraction–repulsion tour of 70K Fashion MNIST images. Middle: UMAP-Validating little PCA tour of 290K single-cell RNA-seq cells. Right: Sequential embedding tour of 3M arXiv titles and abstracts. not a reflection of genuine data structure. Inspecting the images confirms that all 96 points are short trousers whose compact pixel silhouette resembles upper-body garments more than full-… view at source ↗
read the original abstract

Understanding high-dimensional data requires projecting it into lower-dimensional spaces, but any single projection inevitably loses information or introduces distortions. Tours address this limitation through animation of 2D projection sequences, yet existing tools present tradeoffs in the freedom and steerability of projection traversal, providing little to no ability to move between expert-guided paths and unrestrained exploration. We present dtour, a tour interface that combines static projection previews, reversible scrubbing along continuous geodesic projection paths, manual projection manipulation, and a wandering grand tour, all within a single progressive exploration interface. dtour scales to millions of points via GPU-accelerated rendering, runs in any modern browser, and integrates with both Python and JavaScript ecosystems. We demonstrate dtour on text, image, and single-cell data for two usage scenarios: gradually revealing structure in high-dimensional data and validating non-linear dimensionality reduction outputs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper presents dtour, a web-based steerable tour interface for high-dimensional data that integrates static projection previews, reversible scrubbing along continuous geodesic paths, manual projection manipulation, and a wandering grand tour in one progressive interface. It emphasizes GPU-accelerated rendering for millions of points, browser execution, Python/JS integration, and qualitative demonstrations on text, image, and single-cell data for revealing structure and validating non-linear dimensionality reduction.

Significance. If the integrated steerable features deliver the claimed usability and insight gains without new drawbacks, dtour would advance interactive high-dimensional visualization tools, particularly for ML and bioinformatics workflows. The implementation strengths—GPU rendering, browser compatibility, and dual-ecosystem APIs—are concrete practical contributions that could be adopted readily.

major comments (1)
  1. [Abstract and §5 (Demonstrations)] Abstract and demonstration sections: the central claim that combining static previews, geodesic scrubbing, manual control, and grand tour yields practical usability/insight advantages without new drawbacks is unsupported; only qualitative scenario demonstrations are provided, with no user studies, task metrics, A/B comparisons to tourr/GrandTour, or cognitive-load measures, leaving the net-positive assumption untested and load-bearing for the contribution.
minor comments (2)
  1. [Methods/Implementation] The geodesic path computation and scrubbing mechanics are referenced but would benefit from a brief algorithmic outline or pseudocode to aid reproducibility.
  2. [Figures] Figure captions and legends could more explicitly label the four interaction modes (preview, scrub, manual, wander) to help readers map visuals to the described features.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive review and positive assessment of dtour's implementation and practical contributions. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract and §5 (Demonstrations)] Abstract and demonstration sections: the central claim that combining static previews, geodesic scrubbing, manual control, and grand tour yields practical usability/insight advantages without new drawbacks is unsupported; only qualitative scenario demonstrations are provided, with no user studies, task metrics, A/B comparisons to tourr/GrandTour, or cognitive-load measures, leaving the net-positive assumption untested and load-bearing for the contribution.

    Authors: We agree that the manuscript provides only qualitative scenario demonstrations on text, image, and single-cell datasets to illustrate the two usage scenarios (revealing structure and validating non-linear dimensionality reduction). No user studies, quantitative task metrics, A/B comparisons to tools such as tourr or GrandTour, or cognitive-load measurements are included, so the claim of practical usability and insight advantages without new drawbacks rests on illustrative examples rather than empirical validation. This is a genuine limitation for a tool-focused paper. We will revise the abstract and §5 to tone down the language, framing the demonstrations as concrete illustrations of how the integrated features enable progressive exploration rather than as evidence of net-positive effects. We will also add an explicit limitations subsection noting the lack of formal evaluations and identifying controlled user studies as future work. This is a partial revision, as we can clarify the presentation and add discussion but cannot introduce new empirical data. revision: partial

Circularity Check

0 steps flagged

No significant circularity; paper is purely descriptive with no derivations or self-referential claims

full rationale

The paper presents dtour as a software interface combining static previews, geodesic scrubbing, manual manipulation, and grand tours for high-dimensional data exploration. It details GPU rendering, browser integration, Python/JS APIs, and qualitative demos on text/image/single-cell data, but contains no equations, fitted parameters, predictions, uniqueness theorems, or derivation chains. Claims rest on implementation and examples rather than any step that reduces to its own inputs by construction. No self-citation load-bearing arguments or ansatzes are present, making the work self-contained as a tool description.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a software tool paper with no mathematical model, derivations, or postulated entities; the ledger is empty by nature of the contribution type.

pith-pipeline@v0.9.0 · 5442 in / 1055 out tokens · 30443 ms · 2026-05-08T16:55:00.312581+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    Agrawal, A

    A. Agrawal, A. Ali, and S. Boyd. Minimum-distortion embedding. F ound. Trends Mach. Learn., 14(3):211–378, 2021

    work page 2021

  2. [2]

    D. Asimov. The grand tour: A tool for viewing multidimensional data. SIAM J. Sci. Stat. Comput., 6(1):128–143, 1985

    work page 1985

  3. [3]

    Becht, L

    E. Becht, L. McInnes, J. Healy, C.-A. Dutertre, I. W. Kwok, L. G. Ng, F. Ginhoux, and E. W. Newell. Dimensionality reduction for vi- sualizing single-cell data using umap.Nat. Biotechnol., 37(1):38–44, 2019

    work page 2019

  4. [4]

    Belkin and P

    M. Belkin and P. Niyogi. Laplacian eigenmaps for dimensionality reduction and data representation.Neural Comput., 15(6):1373–1396, June 2003

    work page 2003

  5. [5]

    Buja and D

    A. Buja and D. Asimov. Grand tour methods: an outline. InProc. 17th Symp. Interface, 1986

    work page 1986

  6. [6]

    A. Buja, D. Cook, D. Asimov, and C. Hurley. Computational meth- ods for high-dimensional rotations in data visualization.Handbook of statistics, 24:391–413, 2005

    work page 2005

  7. [7]

    J. N. B ¨ohm, P. Berens, and D. Kobak. Attraction-repulsion spectrum in neighbor embeddings.JMLR, 23(95):1–32, 2022

    work page 2022

  8. [8]

    Catmull and R

    E. Catmull and R. Rom. A class of local interpolating splines. In Computer aided geometric design, pages 317–326. Elsevier, 1974

    work page 1974

  9. [9]

    J. M. Chambers, W. S. Cleveland, B. Kleiner, and P. A. Tukey.Graph- ical Methods for Data Analysis. Wadsworth & Brooks, 1983

    work page 1983

  10. [10]

    Chari and L

    T. Chari and L. Pachter. The specious art of single-cell genomics. PLoS Comput. Biol., 19(8):e1011288, 2023

    work page 2023

  11. [11]

    J. Chen, S. Xiao, P. Zhang, K. Luo, D. Lian, and Z. Liu. M3- Embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation.arXiv, 2024

    work page 2024

  12. [12]

    Cook and A

    D. Cook and A. Buja. Manual controls for high-dimensional data projections.Journal of Computational and Graphical Statistics, 6(4):464–480, 1997

    work page 1997

  13. [13]

    D. Cook, A. Buja, J. Cabrera, and C. Hurley. Grand tour and projection pursuit.Journal of Computational and Graphical Statistics, 4(3):155– 172, 1995

    work page 1995

  14. [14]

    Greene, G

    E. Greene, G. Finak, L. A. D’Amico, N. Bhardwaj, C. D. Church, C. Morishima, N. Ramchurren, et al. New interpretable machine- learning method for single-cell data reveals correlates of clinical re- sponse to cancer immunotherapy.Patterns, 2(12), 2021

    work page 2021

  15. [15]

    Harrison

    P. Harrison. langevitour: Smooth interactive touring of high dimen- sions, demonstrated with scrna-seq data.The R Journal, 15(2):206– 219, 2023

    work page 2023

  16. [16]

    Hart and E

    C. Hart and E. Wang.detourr: Portable and Performant Tour Anima- tions, 2022. R package version 0.2.0

    work page 2022

  17. [17]

    Kobak and P

    D. Kobak and P. Berens. The art of using t-sne for single-cell tran- scriptomics.Nat. Commun., 10(1):5416, 2019

    work page 2019

  18. [18]

    La Manno, K

    G. La Manno, K. Siletti, A. Furlan, D. Gyllborg, E. Vinsland, A. Mossi Albiach, C. Mattsson Langseth, I. Khven, A. R. Lederer, L. M. Dratva, et al. Molecular architecture of the developing mouse brain.Nature, 596(7870):92–96, 2021

    work page 2021

  19. [19]

    Lause, P

    J. Lause, P. Berens, and D. Kobak. The art of seeing the elephant in the room: 2d embeddings of single-cell data do make sense.PLoS Comput. Biol., 20(10):e1012403, 2024

    work page 2024

  20. [20]

    S. Lee, D. Cook, N. da Silva, U. Laa, N. Spyrison, E. Wang, and H. S. Zhang. The state-of-the-art on tours for dynamic visualization of high-dimensional data.WIREs Computational Statistics, 14(4):e1573, 2022

    work page 2022

  21. [21]

    S. Lee, U. Laa, and D. Cook. Casting multiple shadows: High- dimensional interactive data visualisation with tours and embeddings. arXiv, 2021

    work page 2021

  22. [22]

    Lekschas

    F. Lekschas. Regl-Scatterplot: A Scalable Interactive JavaScript- based Scatter Plot Library.Journal of Open Source Software, 8(84):5275, Apr. 2023

    work page 2023

  23. [23]

    Lekschas and T

    F. Lekschas and T. Manz. Jupyter scatter: Interactive exploration of large-scale datasets.Journal of Open Source Software, 9(101):7059, 2024

    work page 2024

  24. [24]

    M. Li, Z. Zhao, and C. Scheidegger. Visualizing neural networks with the grand tour.Distill, 2020. https://distill.pub/2020/grand-tour

    work page 2020

  25. [25]

    F. Mair, J. R. Erickson, M. Frutoso, A. J. Konecny, E. Greene, V . V oil- let, N. J. Maurice, A. Rongvaux, D. Dixon, B. Barber, et al. Extri- cating human tumour immune alterations from tissue inflammation. Nature, 605(7911):728–735, 2022

    work page 2022

  26. [26]

    T. Manz, N. Abdennur, and N. Gehlenborg. anywidget: reusable wid- gets for interactive analysis and visualization in computational note- books.Journal of Open Source Software, 9(102):6939, 2024

    work page 2024

  27. [27]

    T. Manz, F. Lekschas, E. Greene, G. Finak, and N. Gehlenborg. A gen- eral framework for comparing embedding visualizations across class- label hierarchies.IEEE Transactions on Visualization and Computer Graphics, pages 1–11, 9 2024

    work page 2024

  28. [28]

    L. McInnes. DataMapPlot: Creating beautiful plots of data maps, 2024

    work page 2024

  29. [29]

    McInnes, J

    L. McInnes, J. Healy, and J. Melville. Umap: Uniform manifold ap- proximation and projection for dimension reduction.arXiv, 2018

    work page 2018

  30. [30]

    Nussbaum and B

    Z. Nussbaum and B. Duderstadt. Training sparse mixture of experts text embedding models.arXiv, 2025

    work page 2025

  31. [31]

    D. Ren, F. Hohman, H. Lin, and D. Moritz. Embedding atlas: Low- friction, interactive embedding visualization, 2025

    work page 2025

  32. [32]

    G. G. Robertson, S. K. Card, and J. D. Mackinlay. Information visual- ization using 3d interactive animation.Communications of the ACM, 36(4):57–71, 1993

    work page 1993

  33. [33]

    Rodrigues, F

    N. Rodrigues, F. L. Dennig, V . Brandt, D. A. Keim, and D. Weiskopf. Comparative evaluation of animated scatter plot transitions.IEEE Transactions on Visualization and Computer Graphics, 30(6):2929– 2941, 2024

    work page 2024

  34. [34]

    Shneiderman

    B. Shneiderman. The eyes have it: A task by data type taxonomy for information visualizations. InThe craft of information visualization, pages 364–371. Elsevier, 2003

    work page 2003

  35. [35]

    Singh, M

    A. Singh, M. D’Arcy, A. Cohan, D. Downey, and S. Feldman. SciRepEval: A multi-format benchmark for scientific document rep- resentations. InProc. Empirical Methods in Natural Language Pro- cessing, 2023

    work page 2023

  36. [36]

    Sivaraman, Y

    V . Sivaraman, Y . Wu, and A. Perer. Emblaze: Illuminating machine learning representations through interactive comparison of embedding spaces. InProceedings of the 27th International Conference on Intel- ligent User Interfaces, pages 418–432, 2022

    work page 2022

  37. [37]

    Smilkov, N

    D. Smilkov, N. Thorat, C. Nicholson, E. Reif, F. B. Vi ´egas, and M. Wattenberg. Embedding projector: Interactive visualization and interpretation of embeddings.arXiv, Nov. 2016

    work page 2016

  38. [38]

    Spyrison and D

    N. Spyrison and D. Cook. spinifex: An R package for creating a manual tour of low-dimensional projections of multivariate data.The R Journal, 12(1):243–257, 2020

    work page 2020

  39. [39]

    Swayne, D

    D. Swayne, D. T. Lang, A. Buja, and D. Cook. Ggobi: Xgobi re- designed and extended. InProceedings of the 33th Symposium on the Interface: Computing Science and Statistics, 2001

    work page 2001

  40. [40]

    D. F. Swayne, D. Temple Lang, A. Buja, and D. Cook. GGobi: Evolv- ing from XGobi into an extensible framework for interactive data visu- alization.Computational Statistics & Data Analysis, 43(4):423–444, 2003

    work page 2003

  41. [41]

    E. R. Tufte.The visual display of quantitative information, volume 2. Graphics press Cheshire, CT, 1983

    work page 1983

  42. [42]

    Van der Maaten and G

    L. Van der Maaten and G. Hinton. Visualizing data using t-sne.JMLR, 9(11), 2008

    work page 2008

  43. [43]

    Z. J. Wang, F. Hohman, and D. H. Chau. WizMap: Scalable Inter- active Visualization for Exploring Large Machine Learning Embed- dings.arXiv 2306.09328, 2023

    work page arXiv 2023

  44. [44]

    Wattenberg, F

    M. Wattenberg, F. Vi´egas, and I. Johnson. How to use t-sne effectively. Distill, 2016

    work page 2016

  45. [45]

    Wickham, D

    H. Wickham, D. Cook, H. Hofmann, and A. Buja. tourr: An R pack- age for exploring multivariate data with projections.Journal of Statis- tical Software, 40(2):1–18, 2011

    work page 2011

  46. [46]

    H. Xiao, K. Rasul, and R. V ollgraf. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms.arXiv, 2017

    work page 2017

  47. [47]

    Zhang, Z

    Z. Zhang, Z. Liao, H. Yu, P. Di, and R. Wang. F2llm-v2: Inclusive, performant, and efficient embeddings for a multilingual world, 2026

    work page 2026