Hacking sklearn's t-SNE implementation to animate embedding process
To generate the animation, just run
source animate.sh
Result on MNIST dataset, early exaggeration for first 250 iters:
MNIST, early exaggeration all the way through, fixed window: