DruidJS

Appearance

DruidJS Showcases ​

Explore the capabilities of DruidJS through our interactive showcases. We've organized these examples into specialized categories to demonstrate different aspects of the library.

Explore by Category ​

  • Standard Projections
    A gallery of classic dimensionality reduction algorithms on the Iris dataset. Include methods like PCA, t-SNE, UMAP, MDS, TriMap, ISOMAP, TopoMap, Sammon, LLE, FastMap, SMACOF, LDA, LSP, LTSA, SQDMDS.

  • Clustering Pipelines
    Learn how to build pipelines using powerful clustering algorithms (including K-Means, OPTICS, CURE, and Hierarchical Clustering) combined with UMAP to project high-dimensional structure.

  • Hierarchical Dendrograms
    A visual interactive representation of the internal recursive nested-cluster logic powering the Hierarchical Clustering algorithm, rendered natively in D3.

  • Interactive Optimization
    Watch iterative algorithms (t-SNE, UMAP, TriMap, Sammon, SMACOF, SQDMDS) in action on synthetic and real datasets, and see how they optimize embeddings over time.

  • Metric Sensitivity
    Discover how the choice of distance metrics (Euclidean, Cosine, Manhattan, Chebyshev) influences the final spatial layout in a Sammon projection.

  • Topological Preservation
    A deep dive into TopoMap and its unique approach to preserving data 0-dimensional topological connectivity on concave datasets like Moons.

  • Earth Mover Analysis
    Compare non-negative distribution histograms (Gaussian, Uniform, Bimodal, Exponential) using the Wasserstein distance metric and MDS to project complex 1D distributions as 2D spatial coordinates.

  • Global Projections
    Geospatial analysis using the Haversine metric on world city coordinates, charting spherical data down to flat manifolds while retaining geographical distances.

  • KNN Image Search
    Fast approximate similarity search (via an HNSW index graph) traversing a subset of the MNIST handwritten digits dataset in its native 784-dimensional space.

  • Automatic Discovery
    Let X-Means automatically find the optimal number of clusters for a randomized blob dataset, dynamically tuning to the Bayesian Information Criterion without manual parameter nudging.

About the Data ​

These showcases use various datasets provided by the @saehrimnir/mistle library, including:

  • Iris: Classic botanical measurements of Iris flowers.
  • Wine: Chemical analysis of wines from a specific region.
  • Penguins: Body measurements of various penguin species.
  • MNIST: High-dimensional representations of pixel-based handwritten digits.
  • Swiss Roll / Moons / Blobs: Synthetic datasets designed to test specific nonlinear manifold structures and clustering characteristics.