mRMR_0.9_compiled.rar_code mRMR_最大最小冗余_最大最小相关_最小冗余特征_特征选择mrmr

********************************************************** To run the mRMR variable/feature selection for discrete variables, you need to add the path of mi_0.9 in your Matlab path. Then simply run mrmr_mid_d.m or mrmr_miq_d.m for your data. If you have continuous data as input, you need to discretize the data first. Simple methods to do this can be thresholding at the mean or mean-plus/minus-std. Of course, you may also try the MI computation for continuous variables directly, as I showed in the my paper (below). However, typically the results are not as good as discrete variables. Note that this version is old and uses double-precision in mutual information computation, thus the feature selection results may be slightly different if you also compare the against those produced by the newer C versions downloadable from our website. The C versions uses single precision for floating numbers to save some memories. The codes cannot be re-distributed without permission from the author, Hanchuan Peng. We hope you cite our work as follows, which you can download the paper at Hanchuan Peng's web site http://research.janelia.org/peng (you may google and find out the latest website). Hanchuan Peng, Fuhui Long, and Chris Ding, "Feature selection based on mutual information: criteria of max-dependency, max-relevance, and min-redundancy," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 27, No. 8, pp.1226-1238, 2005. Should you have any question, please send email to hanchuan.peng@gmail.com or pengh@janelia.hhmi.org . **********************************************************