Demos

A collection of demonstrations of SimilaritySearch.jl and related packages, using real and synthetic datasets. Unlike the package tutorials, these demos use larger datasets and longer computations to produce richer visualisations.

Tip: run Julia with -t auto (or JULIA_NUM_THREADS=auto) to enable multithreading in all demos.

---

Synthetic / Geometric

Conceptually simple datasets that make algorithmic behaviour easy to visualise.

• 2D Synthetic Dataset — SearchGraph on 100k 2D points; batch search, closestpair, near-duplicate detection.
• Prime Number Factors — Integers as prime-factor sets; Dice distance; UMAP; all-kNN recall evaluation.
• All-kNN and Center Selection — allknn, fft, dnet, randsel, multirandsel, neardup; applied to Fashion-MNIST prototypes.

---

Vision / Images

Demos on image datasets: visual search, near-duplicates, and dimensionality reduction.

• MNIST Handwritten Digits — SearchGraph + UMAP; visual kNN search; near-duplicate digit detection.

---

Language / Text

Demos on word embeddings and full-text search.

• Wikipedia LSI Word Embeddings — LSI embeddings from Wikipedia ES/EN; synonym search; analogy test; UMAP projection of the lexicon.
• Wikipedia BM25 Search — High-performance full-text BM25 search with TextSearch 1.1 Inverted Files and query expansion on 60k Wikipedia paragraphs.
• Emoji-Space — Tweet embeddings with TextSearch 1.1 TF-IDF; UMAP.

---

Advanced Operations

Demos of SimilaritySearch operations beyond basic kNN.

• Bichromatic Operations — bichromatic_closestpair, bichromatic_kclosestpairs, bichromatic_metricjoin; visual 2D example + cross-lingual word alignment with Wikipedia embeddings.

---

Manifold Learning

Using SimSearchManifoldLearning.jl with ManifoldLearning.jl.

• Isomap with ApproxEuclidean — S-Curve and prime-gap time series embedded with Isomap using approximate nearest neighbours.