NSB Entropy Estimation

Version 1.14 Differences from 1.13. In the version 1.11, there was a new bug introduced, which prevented execution of the code: the program would not have outputed anything. This bug has been fixed now. The program can now run OK with -c1 and -c2 switches. -c0 (default), which includes full nsb integration, does not work due to some incompatibility with the newer version of the GSL library (at least on my Mac OS X 10.8.4). I will correct this error in the next release. For now, one can use the C++ version to count words in the data, and then the Octave/Matlab version to actually perform integration. *********************** Version 1.13 Differences from 1.11: 1) Compiles with Apple's gcc 4.2.1 for Darwin 12.4.0, on x86_64 2) R-wrapper nsb.entropy.R added written by Jean Hausser in 2008 (took only 5 years to add it!) 3) License changed to GPLv3. ************************ Version 1.11. Differences from 1.1: 1) Compiles with Apple's gcc 4.0.1 for Darwin 8.7.0, PowerPC 2) Compiles on i686 Linux-Debian 3) Parameters for the command line option -c have changed, now allowing to either: (0) do full NSB analysis; (1) calculate frequency of occurence of frequencies of samples; (2) calculate frequency of occurencies of samples. 4) Bug corrected where a nonzero probability caused the shifts to be set to 1. 5) Various small bugs corrected. ************************** Version 1.1 DIFFERENCES from 1.0 1) A new command line option -rY/N has been added, which has the deafult value N. If r==N, the program executes as earlier. If r==Y, then the histogram of the occurences is partitioned into very high (substantial fraction, i.e., over 10% and more that a few thousands of all samples in each bin), high (more than about a hundred samples in each bin), and low (all other) occupancy bins. The NSB entropy for each set is calculated separately, and then the total entropy is evaluated using entropy's additivity (and the error on entropy, of course, has a component related to the imprecision of knowing the occupancy of each class). This option can be used to test for bias in the NSB estimator, since, for unbiased estimation, there should be no difference between direct and partitioned evaluation. 2) Various errors that made the code crash have been removed. 3) The code now supports keeping the necessary spline interpolators for xi(beta) function in memory for more than one alphabet size so that, in particular, -rY option can be used with faster execution times. ********************************************* Installation Instructions: Brief readme, assuming you are running OS X REQUIREMENTS This software routines require several things to be installed: gsl, gmp, octave, gnuplot (from fink), and gcc 3.3 (or higher) (from apple) APPLE: To check which version you are running type foo% gcc --version If you are not running gcc 3.3, log in to http://developer.apple.com/index.html, and goto downloads, then developer tools, and download the August 2003 or later update to gcc, so that you are running gcc version 3.3 or later. A later version should work as well (at the time of the writing of this document, the latest gcc version is 4.0.1). If the only available version of gcc is 2.xx, uncomment the line #define GCC_2 in the include/compiler.h file. FINK (http://fink.sourceforge.net/): (you might want to make sure your fink is up-to-date: foo% fink selfupdate;fink update-all;apt-get update;apt-get upgrade; apt-get dist-upgrade;apt-get clean;apt-get autoclean;apt-get check ) Required via fink are gsl, gmp, octave, gnuplot GSL and GMP are periodically "unstable" modules. This means you need to configure fink so that it has access to these modules; please see Fink documentation for instructions. INSTALL Assuming the software is in the directory nsb-entropy/, which is the current directory: make cd build ./nsb-entropy -h USAGE Type ./nsb-entropy -h to get the full description of program options.