Repository files navigation

PyTorch implementation of the t-stochastic neighbor embedding algorithm described inVisualizing Data using t-SNE.

While CUDA support is not required for this library, the best performance can be achieved when this library is used on a system with CUDA support.

Requires Python 3.7 or later

pip3 install tsne-torch

git clone https://github.com/palle-k/tsne-pytorch.gitcd tsne-pytorchpython3 setup.py install

fromtsne_torchimportTorchTSNEasTSNEX= ...# shape (n_samples, d)X_emb=TSNE(n_components=2,perplexity=30,n_iter=1000,verbose=True).fit_transform(X)# returns shape (n_samples, 2)

python3 -m tsne_torch --xfile<path> --yfile<path>

This is our result compared to the result of the author's Python implementation on a subset of the MNIST dataset:

• PyTorch result

• python result

This code highly inspired by

• author's python implementation codehere.