Pytorch implementation of Kohonen's Self-Organizing MAP.
- Fast direct updates with mini-batch on GPU (using optim is slow)
- Initialize map center with centroid of data
This code does not require nunif library.
python3 -m playground.som.train_mnist
The generated images are saved in tmp/som
.
Autoencoder based Deep Convolutional SOM that I designed at random.
python3 -m playground.som.train_aesom_mnist
It's less blurry.