This archive contains methods for assessing data mining results on matrices using randomization as described in Markus Ojala: Assessing Data Mining Results on Matrices with Randomization. In ICDM'10: Proceeding of the 10th IEEE International Conference on Data Mining, pp. 959-964. ------- General ------- The SwapConstrained method is implemented with Java but it is intended to use with the provided Matlab interface. Use the following commands in Matlab to get help using the methods: help swap %the main randomization method help discretize %the method controlling discretization Short example: X = rand(1000,100); DX = discretize(X); %discretized version of X DY = swap(DX); %randomized discretization Y = undiscretize(DY); %randomized version of X ---------------------- Descriptions of files ---------------------- MAIN METHODS - discretize.m discretizes matrix X - swap.m randomizes discretization DX - undiscretize.m undiscretizes DY ANALYSIS METHODS - calcRankDiff.m difference between two discretized matrices - calcRankDist.m distance between two discretized matrices - calcRankError.m average CDF error between two discretized matrices JAVA PACKAGES - swap.jar java classes of SwapConstrained (swap.m needs this) - trove-2.1.0.jar auxiliary package used by swap.jar, taken from http://trove4j.sourceforge.net/ JAVA SOURCES - Swap.java SwapConstrained for full matrices - SwapParse.java SwapConstrained for sparse matrices - MersenneTwisterFast.java Taken from http://www.cs.gmu.edu/~sean/research/mersenne/MersenneTwisterFast.java ----------------------------------- Note about Matlab and Java versions ----------------------------------- The methods works on such with Matlab 7.5 and newer. With older Matlab versions, Matlab has to be changed to use newer Java virtual machine (at least 1.5), see Matlab support in http://www.mathworks.com/support/solutions/data/1-1812J.html for more help. To increase the heap space for the JVM, see http://www.mathworks.com/support/solutions/data/1-18I2C.html for more help.