Skip to content
/ CFGAN Public

Implementation of CFGAN CFGAN: A Generic Collaborative Filtering Framework based on Generative Adversarial Networks

License

Notifications You must be signed in to change notification settings

ARASC/CFGAN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CFGAN

Implementation of CFGAN. The implementation based on Pytorch is done, and the max precision on ML100k is 0.49 in 1000 train epochs, when use the paramaters in the paper.

cfgan.py : the model of CFGAN

data.py : load data

main.py : train and show result

Run the demo :

Python main.py

You can modify paramaters in main.py, for example, modify the alpha to 0.5(in fact, result turn better when you do that)

Next, we consider implementing another version based on Tensorflow.

Author: Xuxin Zhang, HUST, [email protected]

Reference: Chae D K , Kang J S , Kim S W , et al. CFGAN: A Generic Collaborative Filtering Framework based on Generative Adversarial Networks[C]// the 27th ACM International Conference. ACM, 2018.

Baseline: https://github.com/1051003502/CFGAN

About

Implementation of CFGAN CFGAN: A Generic Collaborative Filtering Framework based on Generative Adversarial Networks

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published