eyounx / GANMM

Mixture of Generative Adversarial Networks for Clustering

GANMM

GANMM code for the paper

Yang Yu, Wen-Ji Zhou. Mixture of GANs for clustering. In: Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI'18), Stockholm, Sweden.

Requirement

• python 3.5
• argparse
• pickle
• tensorflow(tested with GPU version) == 1.0.0
• numpy == 1.12.1
• sklearn == 0.18.1

Files

• GANMM.py implements the algorihtm
• main.py is the demo that uses GANMM to cluster some data sets as in the paper
• Data contains data sets
• nets network structure
• tflib tensorflow components (modified from https://github.com/igul222/improved_wgan_training)

Run Demo

• To run experiment on mnist data set, just running:

# mnist raw data
python main.py mnist

# mnist preprocessed by stacked auto-encoder
python main.py sae_mnist

GPU version of TensorFlow is recommended for mnist data set. Tensorflow (CPU) may not support data_format="NCHW" in Conv2DBackpropFilter operate.

• Two UCI-dataset:

# Image Segmentation data set
python main.py seg

# Artificial Characters data set
python main.py chara

• On different data scale:

python main.py seg --scale 0.5