Awesome

DeepEmbeddedClustering

chainer implementation of Deep Embedded Clustering(Unsupervised Deep Embedding for Clustering Analysis)
In this code, we use MNIST as training data.

Requirement

• Chainer 2.0.0
• Cupy 1.0.0
  • if use GPU
• scikit-learn 0.18.1

Running

Pretraining

$ python pretraining.py --gpu=0 --seed=0 

--gpu=0 turns on GPU. If you turn off GPU, use --gpu=-1 or remove --gpu option. --seed=0 means random seed.

Training model

$ python main.py --gpu=0 --seed=0 --model_seed=0 --cluster=10 

--gpu and --seed means same as before. --model_seed is seed number when pretraing.
Every five iteration, save embedding result in directory like modelseed0_seed0/.
I used t-SNE and compress embedding vector to 2-dim. And I saved embedding result of 500 data as scatter plot.

Reference

Junyuan Xie, Ross Girshick, Ali Farhadi, "Unsupervised Deep Embedding for Clustering Analysis" https://arxiv.org/abs/1511.06335