Release History

This contains the release notes for the current version (2.0.0a32) of nptsne.

Release notes

Version 2.0.0

Supports python 3.10, 3.11, 3.12, 3.13, 3.14 on Windows, Ubuntu (using manylinux_2_28) and Macos (restricted to arm64).

Changelog 2.0.0

  • Using the latest HDILib iteration that gives cross-platform Vulkan driver support.

  • Vulkan drivers (or other drivers) can be selected on the tSNE instance.

  • The Vulkan driver is now the default on MacOS giving improved performance.

  • tSNE over the Vulkan driver may be faster for 10^5 points and less. For very large numbers of points the OpenGL driver (with compute shader) is about 20% faster.

Previous versions

Version 1.3.0

Supports python 3.9, 3.10, 3.11, 3.12, 3.13 on Windows, Ubuntu (using manylinux_2_28) and Macos (restricted to arm64).

Changelog 1.3.0

  • Using the latest HDI Library v1.2.10 supporting the additional Annoy knn method.

  • The demos at Basic HSNE demo code have been updated to include more exact requirements specification and have been updated for the newer matplotlib and PyQt6 apis. See Demo list

Version 1.2.0

Supports python 3.6, 3.7, 3.8, 3.9 on Windows, Ubuntu (using manylinux2010) and Macos.

Changelog 1.2.0

  • Using the latest HDI Library v1.2.2 supporting the additional Annoy knn method.

  • Added KnnDistanceMetric to the TextureTsne and TextureTsneExtended classes

  • Added a fast, non-exact implementation of the algorithm to get an area of interest from a landmark based selection at nptsne.hsne_analysis.Analysis.get_fast_area_of_influence(). The algorithm pre-calculates a set of non-overlapping AOIs for each landmark at an HSNE scale. For the correct AOI (according to the [NP2019]) use the slower nptsne.hsne_analysis.Analysis.get_area_of_influence().

  • An additional demo set (DC Mall) has been added to the data and the Extended HSNE demo code was updated to allow the display of a background image. See Demo list

  • nptsne has added typing as specified in PEP 484. The demos have also been extended to use PEP 484. Types are included in the package using __init__.pyi files as specified in PEP 561.

Version 1.1.0

Supports python 3.6, 3.7, and 3.8 on Windows, Ubuntu (using manylinux2010) and Macos.

Changelog 1.1.0

  • Using the latest HDI Library v1.2.1 supporting the additional Annoy knn method.

  • Building with cibuildwheel to give manylinux support for a wide range of linux platforms.

  • This version is extended with HSNE support. This comprises the HSne class which can generate or load an hsne analysis and supporting classes (Analysis and AnalysisModel) that can be used to navigate the HSne hierarchy.

  • Additionally a number of demos of GPU accelerated t-SNE and HSNE are available at Demo list

Version 1.0.0

Supports python 3.6 and 3.7 Windows, Ubuntu (xenial/16.04 and upward) and Macos.

Changelog 1.0.0

Using the latest HDI Library which supports both GPU computer shader (Windows and Linux) and GPU rasterizer (Macos) versions of the underlying tSNE algorithm.

Added reinitialize_transform to TextureTsneExtended to allow quick embedding regeneration.

Version 0.2.0

QT dependency removed. Windows release only

Changelog 0.2.0

This was a Windows only release without QT otherwise it is unchanged from the previous version

Version 0.1.1a1

This is an alpha release of the nptsne library. As such the API may change in future releases based on user feedback.

The initial alpha release is being distributed to a limited audience via GitHub. The current roadmap is to switch to PyPi when the API is declared stable.

Changelog 0.1.1a1

This is the first release of the nptsne package.