pavlin-policar/openTSNE

Extensible, parallel implementations of t-SNE

What it solves

It provides a high-performance Python implementation of t-Distributed Stochastic Neighbor Embedding (t-SNE), a dimensionality-reduction algorithm used to visualize high-dimensional data sets. It addresses common t-SNE limitations by improving speed and scalability, allowing the visualization of millions of data points, and enhancing the preservation of global data alignment.

How it works

openTSNE implements two efficient algorithms for t-SNE: FIt-SNE (the default) and Barnes-Hut. It utilizes C/C++ binaries and supports OpenMP for multi-threading. For maximum performance on large datasets, it can integrate with FFTW3 to accelerate Fast Fourier Transforms.

Who it’s for

Data scientists and researchers, particularly those working with single-cell transcriptomics or other massive high-dimensional datasets that require clear visual clusters.

Highlights

  • Scalable to millions of data points.
  • Ability to add new data points to existing embeddings.
  • Modular Python design.
  • Support for multiple acceleration algorithms (FIt-SNE and Barnes-Hut).
  • Techniques to improve global alignment of visualizations.

Related

  • Project
    CannyLab/tsne-cuda

    A GPU-accelerated implementation of the FIt-SNE algorithm that provides massive speedups for visualizing high-dimensional data compared to CPU-based methods.

  • Project
    scikit-tda/kepler-mapper

    A Python implementation of the TDA Mapper algorithm used for visualizing the topological structure of high-dimensional data.

  • Project
    eamid/trimap

    A dimensionality reduction method that uses triplet constraints to preserve the global structure of high-dimensional data, offering a better global view than t-SNE or UMAP.

  • Project
    lmcinnes/umap

    A fast non-linear dimension reduction library for visualizing high-dimensional data and preprocessing for machine learning tasks.

  • Project
    ContextLab/hypertools

    A Python toolbox for dimensionality reduction and visual exploration of high-dimensional data, enabling users to reduce and plot complex datasets in a single call.