A simple alternative implementation of Laurens van der Maaten t-Distributed Stochastic Neighbor Embedding (t-SNE) , with use of kernals.
-
joint probability pij
-
joint probability qij
-
Cost
-
gradient
-
gradient update
for more information read the PDF file included.
Apart from pca, available kernels are:
- iquad
- cauchy
- fourier
- rbf
- poly
- cosine
- anova
you can add new/own kernel and test it, make sure to tune the parameters in f_opts to get better results!
License : MIT and you are free to do what ever you want with it !
