https://github.com/sadit/simsearchmanifoldlearning.jl

Non-linear dimensional reduction using SimilaritySearch (ManifoldLearning and UMAP)
https://github.com/sadit/simsearchmanifoldlearning.jl

Last synced: 3 months ago
JSON representation

Non-linear dimensional reduction using SimilaritySearch (ManifoldLearning and UMAP)

• Host: GitHub
• URL: https://github.com/sadit/simsearchmanifoldlearning.jl
• Owner: sadit
• License: other
• Created: 2022-03-19T16:45:00.000Z (about 3 years ago)
• Default Branch: main
• Last Pushed: 2024-01-10T02:40:27.000Z (over 1 year ago)
• Last Synced: 2025-01-19T05:57:33.084Z (4 months ago)
• Language: Julia
• Homepage:
• Size: 2.68 MB
• Stars: 9
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
[![Dev](https://img.shields.io/badge/docs-dev-blue.svg)](https://sadit.github.io/SimSearchManifoldLearning.jl/dev/)

[![Build Status](https://github.com/sadit/SimSearchManifoldLearning.jl/workflows/CI/badge.svg)](https://github.com/sadit/SimSearchManifoldLearning.jl/actions)

# SimilaritySearch and ManifoldLearning (and UMAP)

This package provides some support to use `SimilaritySearch` with manifold learning methods. In particular,

we implement the required methods to implement `knn` function for `ManifoldLearning` and also provides an `UMAP`

implementation that takes advantage of many `SimilaritySearch` features like multithreading and data independency; it supports string, sets, vectors, etc. under diverse distance functions.

The `ManifoldLearning` support is limited to some structure specification due to the design of the package. See the `ManifoldKnnIndex` type in the documentation pages.

# UMAP implementation

This package also provides a pure Julia implementation of the [Uniform Manifold Approximation and Projection](https://arxiv.org/abs/1802.03426) dimension reduction algorithm

> McInnes, L, Healy, J, Melville, J, *UMAP: Uniform Manifold Approximation and Projection for

> Dimension Reduction*. ArXiV 1802.03426, 2018

The implementation in this package is based on the [UMAP.jl](https://github.com/dillondaudert/UMAP.jl) package by Dillon Gene Daudert and collaborators. Forked and adapted to work with `SimilaritySearch` and take advantage of multithreading systems.

The provided implementation partially supports the `ManifoldLearning` API using `fit` and `predict` and similar arguments. It can use any distance function from `SimilaritySearch`, [Distances.jl](https://github.com/JuliaStats/Distances.jl), [StringDistances.jl](https://github.com/matthieugomez/StringDistances.jl), or any distance function implemented by the user.

Additionally,  it improves multithreading support in the KNN and the UMAP projection.

## Examples and demonstrations

[https://sadit.github.io/SimilaritySearchDemos/](https://sadit.github.io/SimilaritySearchDemos/)

NOTE: Currently, these examples are working with direct implementations of this package. This will be changed soon (I will start once the package becomes part of the general registry).

## Disclaimer

This implementation is a work-in-progress. If you encounter any issues, please create

an issue or make a pull request.