Installing and Using¶
This release supports 3 python versions (3.6, 3.7 and 3.8) on three platforms: Windows, Ubuntu, Macos
On Windows and Linux acceleration is performed using compute shaders. On Macos, due to the lack of compute shader support, tSNE uses a rasterized implementation with a lower performance.
Installing¶
Windows,MacOS,Linux: install from PyPi using: pip install nptsne. The PyPi page.
Demo list¶
A number of demos have been created to help you exploit the accelerated tSNE and HSNE offered by this package. The demos are available in a single demos.zip file.
Demo |
Description |
|---|---|
A complete demonstration including
three different datatypes:
* Image is datapoint (MNIST)
* Pixel is data point (Hyperspectral solar images)
* Multi-dimensional plus meta data (Genetic data)
|
|
Louvain clustering applied to
levels in the HSNE hierarchy
|
|
GPU accelerated t-SNE
on 70000 MNIST points
|
|
GPU accelerated t-SNE
on 70000 MNIST points
with intermediate results
|
|
Run the internal doctest examples in nptsne
Can be used for install verification
|
|
A Jupyter notebook demonstration
of the tSNE API.
Illustrates the following options:
* a plain tSNE
* a pre-loaded embedding
* controlling the iteration when the
exaggeration factor is removed.
|