# CannyLab/tsne-cuda

GPU Accelerated t-SNE for CUDA with Python bindings

Repository: https://github.com/CannyLab/tsne-cuda
Canonical: https://ross.abutalabs.com/products/tsne-cuda
Language: Cuda
License: BSD-3-Clause
License Family: permissive
Topics: cuda, gpu, mnist, tsne, tsne-algorithm, data-visualization, data-analysis, barnes-hut, barnes-hut-tsne, fit-tsne, multithreading, tsne-cuda, python
Last push: 2026-07-22T18:16:49+00:00

## Health v2 (maintenance only)
Score: 95/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 93, release rhythm 94, longevity 100
- inputs: {"age_days": 3085, "days_push": 42, "days_rel": 42, "gap_med": 0, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1957, forks 138 (observed 2026-08-28T04:05:59.162576+00:00)

## What it is
A CUDA-accelerated implementation of the FIt-SNE t-SNE algorithm with Python bindings, offering up to 1200x speedup over scikit-learn. It exposes an sklearn-compatible API for fast dimensionality reduction on large datasets.

## Use cases
- visualize mnist embeddings with t-SNE on gpu
- fast t-SNE for millions of points
- dimensionality reduction of high-dimensional embeddings
- cluster visualization of cifar-10 features
- speed up sklearn TSNE with cuda
- barnes-hut t-SNE in python

## When to choose
- you need t-SNE on large datasets (100k+ points) and have an NVIDIA GPU
- you want an sklearn-compatible TSNE drop-in replacement
- embedding quality comparable to reference implementations matters but runtime must be seconds

## When to avoid
- you have no CUDA-capable GPU
- you need UMAP or other manifold methods
- you work only on CPU or with tiny datasets where sklearn suffices

## Facets
- artifact type: library
- maturity: stable
- function: machine-learning, data-visualization, gpu-computing
- domain: machine-learning, data-visualization, data-science
- platform: windows, python
- tags: tsne, cuda, dimensionality-reduction, barnes-hut, python-bindings, sklearn-api, linux, gpu

## Member repositories
- CannyLab/tsne-cuda (main) score 95

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:59.162576+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:06:00.264108+00:00, confidence not recorded.
  - readme: https://github.com/CannyLab/tsne-cuda (fetched 2026-08-28T04:05:59.162576+00:00, sha c939cf1a368a)
- Data as of 2026-08-30T08:39:29.467469+00:00.