/Deep-Clustering-Network

PyTorch implementation of "Towards k-means-friendly spaces: Simultaneous deep learning and clustering," Bo Yang et al., 2017.

Primary LanguagePython

DCN: Deep Clustering Network

Forked from guenthereder (https://github.com/guenthereder/Deep-Clustering-Network)

Results on Mnist dataset

NMI ARI parameters
0.841 0.747 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.005 --lr 0.002
0.800 0.689 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.005 --lr 0.001
0.800 0.684 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.004 --lr 0.001
0.793 0.676 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.01 --lr 0.001
0.758 0.647 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.05 --lr 0.001
0.748 0.629 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.05
0.737 0.618 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.05 --lr 0.0001
0.737 0.595 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.05 --lr 0.0005
0.701 0.581 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50
0.661 0.472 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.001 --lr 0.001
0.627 0.412 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.05 --lr 0.005
0.678 0.526 mnist.py --latent-dim 10 --epoch 50 --pre-epoch 50 --lamda 0.1 --lr 0.001
0.536 0.205 mnist.py --latent-dim 3 --epoch 50 --pre-epoch 50 --lamda 0.005 --lr 0.001
----- ----- ------------
0.81 0.73 original paper claim: pre-/eps 50, lamda 0.05, 4-layer 500-500-2000-10