RCMREG  
Description
C program to compute the estimates of regression coefficients based on lift rank covariance matrix (LRCM), where the multivariate
rank concepts are based on the Oja criterion function.
Getting started:
In Linux/Unix systems, the following instructions should work without any additional software.  The file rcmreg.tar contains all the C source files and the makefile, so the following steps will get you started:
  1. Type tar xvf rcmreg.tar  (this will extract the files in rcmreg.tar to the current working directory
  2. Type make (this will produce the C-executable rcmreg.o  to the current working directory
  3. Type rcmreg and the progam asks all the inputs. Alternatively you can give the required inputs to the program as arguments in the command line e.g. as follows rcmreg -f  mydata.dat -n 25 -k 4 -y 2-b 200
Please consult the manual page for more details. 
R - program (or C interface to R) 
The R-function rcmreg which you can find here is used to access the C routine. The tarred file rcmregR.tar contains all the C-source files to build up the shared library, which is called by the R-function rcmreg. The following instructions should work as is (in the Linux/Unix systems that has R installed):.
  1. Type tar xvf rcmregR.tar (this will untar the files in rcmregR.tar to the current working directory)
  2. Type R CMD SHLIB rcmreg.c nr.c -o extojarnk.so (This will produce the shared library extojarnk.so which is used by the R-function rcmreg)
  3. Start R and  type source("rcmreg.R")  to load the routine rcmreg to your R session
  4. Type rcmreg(mydata,dx)  to calculate the LRCM estimates of the regression coefficients where the first 'dx' columns of the dataset 'mydata' should contain the regression variables (and the remaining columns  the response variables).
Reference
E. Ollila, H. Oja and V. Koivunen : "Estimates of regression  coefficients based on lift rank covariance matrix," Journal  of the American Statistical Association, vol. 98, no. 461,  pp. 90-98, 2003.  PDF
Any comments, bug reports, suggestions can be sent to Esa Ollila, esollila AT wooster DOT hut DOT fi
Back to my homepage