mRMR (minimum Redundancy Maximum Relevance Feature Selection)

[ Frequently Asked Questions | Online Version | C/C++ Version with Source Codes | Matlab Version | Sample Data Sets | Major Publications | BIBM'07 Tutorial Slides]

Online Version: You can run the program using your own data through the following form, -- you can also download the program and run on your own machines (see links below).

* Data file (Standard CSV file format, where each row is a sample and each column is a variable/attribute/feature. MAKE SURE YOUR DATA IS SEPARATED BY COMMA, BUT NOT BLANK SPACE OR OTHER CHARACTERS!! The first row must be the feature names, and the first column must be the classes for samples. You may download a testing example data set here, which is microrray data of lung cancer (7 classes). The data has been discretized as 3-states. Note that the web-based program can only accept a data file with the maximum size 2M bytes, and maximum number of variables = 10000 -- if you have a larger data set, you should download the program and run on your own machine (see download links below).

* What is the feature selection scheme you want to use:

* How many features you want to select:

* What is the type of your data? "Categorical" means each attribute/variable/feature in your data is discretized as a few categorical states. "Continuous" means these attributes take numerical values. If you choose "Categorical" then the last option below will have no effect. For mutual information based feature selection methods like this web-version of mRMR, you might want to discretize your own data first as a few categorical states, -- empirically this leads to better results than continuous-value mutual information computation. You can also use the option below to discretize your data using two thresholds mean+/-k*std.
Categorical
Continuous

* In case you have continuous variables in your data file, what is the threshold you want to discretize your data, i.e. mean +/- k*std:

Sample Data Sets: More microarray gene expression testing data sets:

Discretized NCI data (9 cancers, discretized as 3-states)

Discretized Lung Cancer data (7 subtypes, discretized as 3-states)

Discretized Lymphoma data (9 cancer-subtypes, discretized as 3-states)

Discretized Leukemia data (2 subtypes, discretized as 3-states)

Discretized Colon Cancer data (2 subtypes, discretized as 3-states)

C/C++ Binary Version: You can download several versions and run on your own machine (you can type the program name and get the help information):

Redhat Linux 64-bit-machine version

Redhat Linux 32-bit-machine version

Mac OS X version for PowerPC

Mac OS 10.5 (Leopard) with Intel CPU

Windows XP/Vista 32bit version

C/C++ Source Codes: You can compile the executable for your own machine if you have a GNU C++ compiler. Just check the file "How_to_compile" for details after you download and unzip. For Linux and Mac, this is simply a "make" command; for Windows, you can use the "make" command in MinGW and MSys:

Source code zip file

Matlab Version: I put online again the original Matlab version as many people asked for it. The path of mutual information computation toolbox included in this release needs to be set. For more information see the respective Readme file or the header of a file.

* Matlab version (*.zip) with detailed explanation of input arguments, but without the mutual information toolbox
* Matlab version (*.zip) with the precomplied mutual information toolbox for Windows, Linux and Mac (no C/C++source codes)
* Mutual information toolbox (precompiled for Windows, Linux and Mac) and C/C++ source codes which can be compiled for other operation systems

Major Publications: For more information of mRMR, you may want to read one of the following papers. Details information of more related publications can be found here.

"Feature selection based on mutual information: criteria of max-dependency, max-relevance, and min-redundancy,"

Hanchuan Peng, Fuhui Long, and Chris Ding
IEEE Transactions on Pattern Analysis and Machine Intelligence,
Vol. 27, No. 8, pp.1226-1238, 2005. [PDF]

"Minimum redundancy feature selection from microarray gene expression data,"

Chris Ding, and Hanchuan Peng,
Journal of Bioinformatics and Computational Biology,
Vol. 3, No. 2, pp.185-205, 2005. [PDF]

(A conference version with a different set of results, but the same title, also appeared on:
Proc. 2nd IEEE Computational Systems Bioinformatics Conference (CSB 2003),
pp.523-528, Stanford, CA, Aug, 2003. [PDF])

(Names of Genes selected.)

Go Back to Hanchuan Peng's web site.

Site Meter