AAAEEEE/DFKM

Corresponding Paper

This project corresponds to the paper

Rui Zhang, Xuelong Li, Hongyuan Zhang, and Feiping Nie, "Deep Fuzzy K-Means with Adaptive Loss and Entropy Regularization," IEEE Transactions on Fuzzy Systems, DOI:10.1109/TFUZZ.2019.2945232.

which has been accepted in Sep, 2019.

Author of Code

Hongyuan Zhang and Rui Zhang

Dependency

Now, codes of DFKM implemented by pytorch is available:

For some reasons, the **version-1 is implemented without help of any frameworks like

pytorch-1.3.1
numpy
scikit-learn
scipy

Brief Introduction

DFKM.py: the main source code of DFKM.
data_loader.py: load data from matlab files (*.mat).
utils.py: functions used in experiemnts.
metric.py: codes for evaluation of clustering results.

Samples to run the code is given as follows

import data_loader as loader
data, labels = loader.load_data(loader.USPS)
data = data.T
for lam in [10**-3, 10**-2, 10**-1, 1]:
	print('lam={}'.format(lam))
	dfkm = DeepFuzzyKMeans(data, labels, [data.shape[0], 512, 300], lam=lam, gamma=1, batch_size=512, lr=10**-4)
	dfkm.run()

In fact, the data_loader.py is not necessary. You just need to input a numpy-matrix (n * d) into DeepFuzzyKMeans. If you have any question, please email hyzhang98@gmail.com.

Directory v0

To verify the derivations in our paper, we implement the code of DFKM only by numpy, and the related codes are put into v0(without dl-framework). However, the codes are not clear enough, and they are hard to maintain and update. So we now rewrite the core codes of DFKM.

Citations

@ARTICLE{DFKM,
  author={R. {Zhang} and X. {Li} and H. {Zhang} and F. {Nie}},
  journal={IEEE Transactions on Fuzzy Systems}, 
  title={Deep Fuzzy K-Means with Adaptive Loss and Entropy Regularization}, 
  year={2020},
  volume={28},
  number={11},
  pages={2814-2824},
}

Thanks

Thanks to Xi Peng, Jiashi Feng, Shijie Xiao, Wei-Yun Yau, Joey Tianyi Zhou, and Songfan Yang, "Structured AutoEncoders for Subspace Clustering", IEEE Transactions on Image Processing, vol. 27, no. 10, pp.5076-5086, 2018.

The codes they provide are used in our project.

AAAEEEE / DFKM