Demo code of our AAAI 2016 paper "Random Mixed Field Model for Mixed-Attribute Data Restoration".
%% Code authors: Qiang Li
%% Release time: Jun. 3rd, 2018
%% Current version: RMF_code_v1
- Add path.
First of all, run 'AddPath.m' to add paths of required packages.
- Run the Main code.
There are three demos, i.e.,
'DemoClassifyMissNoisyData.m' gives classification comparison.
'DemoInferMixedNet.m' presents variational inference for data denoising. In this case, the RMF model parameters are user-defined and fixed.
It also gives the MSE plots and graph visualization. Note there maybe cases when denoising cont/disc nodes is ineffective. This is possibly due to a compromise between continuous and discrete nodes.
'DemoLearnStruct.m' presents structure and parameter learning. In this demo, the RMF model is learned using Jason Lee's code.
- Datasets and Corrupted data preparation.
The experiment is conducted on several UCI datasets.
https://archive.ics.uci.edu/ml/datasets.html
Run 'GenMissNoisyData.m' to get the corrupted data.
- Dependencies.
UGM at http://www.di.ens.fr/~mschmidt/Software/UGM_2009.zip
TFOCS at http://tfocs.stanford.edu
MGM at http://www-bcf.usc.edu/~lee715/syntheticExp/syntheticExp.zip
%% Reference noticement:
If you have used the code, please cite the following paper:
[1] Random Mixed Field Model for Mixed-Attribute Data Restoration
Qiang Li, Wei Bian, Richard Yi Da Xu, Jane You and Dacheng Tao
AAAI Conference on Artificial Intelligence (AAAI), Feb. 2016, pp. 1244--1250.
%% Supporting information:
If any questions and comments, feel free to send your email to
Qiang Li (leetsiang.cloud@gmail.com)