For my (Parttime) Master CS thesis project about deep learning I forked BHTSNE to make changes.
This Fork adds:
- (In my opinion) better Python wrapper. It is a 'drop in replacement' following the original tsne interface specifications.
What it does:
tsne.py
reads MNIST(subset) and plot a figure.
Switch between Original TSNE and BH-SNE by commenting out OriginalTSNE
or BHTSNEDropInReplacement
in tsne.py
:
#from OriginalTSNE import tsne
from BHTSNEDropInReplacementTSNE import tsne
I removed the PCA dim. reduction from the original TSNE.
- Changes in C-code: parameters are printed to stdout (some parameters are now printed twice but it works for debugging)
TODO: Generated intermediate files are not being removed.