kokikwbt / dcmf

Python implementation of Dynamic Contextual Matrix Factorization (SDM'15)

DCMF: Dynamic Contextual Matrix Factorization

Unofficial Python implementation of the DCMF algorithm:
Y. Cai, H. Tong, W. Fan, and P. Ji. Fast Mining of a Network of Coevolving Time Series. In SDM, 2015.

DEMO

Please RUN:

python3 dcmf.py --demo

It will start processing the DCMF algorithm over a toy dataset with a synthetic contextual matrix. The results will appear in out directory.

Comand line options

• --output_dir: path of output directory (default: 'out')
• --n_components: dimension of latent variables (default: 2)
• --alpha: weight of contextual infromatioin (default: 0.1)
• --max_iter: maximum number of the EM algorithm (default: 100)
• --tol: tolerance for early stopping (default: 0.1)

Citation

@inproceedings{cai2015fast,
  title={Fast mining of a network of coevolving time series},
  author={Cai, Yongjie and Tong, Hanghang and Fan, Wei and Ji, Ping},
  booktitle={Proceedings of the 2015 SIAM International Conference on Data Mining},
  pages={298--306},
  year={2015},
  organization={SIAM}