Skip to content

Latest commit

 

History

History
39 lines (28 loc) · 1.12 KB

README.md

File metadata and controls

39 lines (28 loc) · 1.12 KB

t-SNE pytorch Implementation with CUDA

pytorch implementation of algorithm described in Visualizing Data using t-SNE. This code support cuda accelerating.

How to use it

Just download the repository, and the unzip mnist2500_X.zip or put feature file and labels file with code

1. run without cuda support

python tsne_torch.py --xfile mnist2500_X.txt --yfile mnist2500_labels.txt --cuda 0

2.run with cuda support

python tsne_torch.py --xfile mnist2500_X.txt --yfile mnist2500_labels.txt --cuda 1

Requirements

  • pytorch
  • matplotlib, numpy

Example

This is our result compare to result of python implementation.

  • pytorch result
  • use time 352s on 2080Ti GPU

computing time pytorch result

  • python result
  • use time 634s on CPU

computing time python result

Credit

This code highly inspired by

  • author's python implementation code here.