NAME
INSTALL Installation instructions for SenseClusters
SYNOPSIS
If you have su or sudo access, you should be able to install and test
the installation of SenseClusters via automatic download from CPAN as
follows:
# install SenseClusters and all dependent CPAN modules
perl -MCPAN -e 'install Bundle::Text::SenseClusters';
# install cluto and SVDPACKC (included in SenseClusters)
cd ~/.cpan/build/Text-SenseClusters-[insert_version]
cd External
csh ./install.sh /usr/local/bin
cd ~
# run SC test cases (note that location of cpan build
# directory might vary on your system.
cd ~/.cpan/build/Text-SenseClusters-[insert_version]
cd Testing
csh ./ALL-TESTS.sh
cd ~
This assumes that /usr/local/bin is in your PATH and is your preferred
location for user installed executable scripts. If it is not, substitute
your preferred directory here.
INSTALLATION OVERVIEW
SenseClusters consists of the core SenseClusters programs (primarily .pl
programs found in this distribution), Perl modules available from CPAN,
and two external programs (SVDPACKC and Cluto). SenseClusters requires
Perl 5.6.0 or better.
SVDPACKC is distributed as C source code, and Cluto is distributed in
binary form for Linux and Solaris. This dependence on CLUTO limits
SenseClusters to running on Linux or Solaris. There is a Windows version
of CLUTO available, but we have not tested how well this integrates into
SenseClusters.
AUTOMATIC INSTALLATION
You may be able to download and install the SenseClusters core,
dependent CPAN modules, and External programs via a single command using
the CPAN module.
If you have sudo or su access, then installation of the CPAN modules and
the core SenseClusters programs can be achieved as follows:
perl -MCPAN -e 'install Bundle::Text::SenseClusters';
This command will install SenseClusters and all dependent CPAN modules.
SenseClusters includes a script that will help you install the External
code as well. You may be able to install the external programs Cluto and
SVDPACKC via the following :
cd ~/.cpan/build/Text-SenseClusters-[insert_version]
cd External
csh ./install.sh INSTALLDIR
cd ~
If you have sudo or su access, INSTALLDIR should be a directory in your
PATH, such as /usr/local/bin. If you do not, you will need to install
into a directory you have read and write access to, and then include in
your path. If install.sh fails for some reason, you will need to install
Cluto and SVDPATH manually, as described below.
At present SenseClusters does not utilize 'make test', so testing must
be done via scripts found in the /Testing directory. Please make certain
to run these tests after the installation of the External programs, CPAN
modules, and SenseClusters core has concluded.
cd ~/.cpan/build/Text-SenseClusters-[insert_version]
cd Testing
csh ./ALL-TESTS.sh
cd ~
If this automatic install does not work, you can download SenseClusters
manually from CPAN <http://search.cpan.org/dist/Text-SenseClusters> or
Sourceforge <http://senseclusters.sourceforge.net> and install as
described in the rest of this document.
SENSECLUSTERS COMPONENTS
CPAN Modules
SenseClusters depends on a number of different CPAN modules. These are
all included in the Bundle above, and can be installed via the Bundle
(recommended) or individually (described below).
* PDL (Perl Data Language, version 2.4.1 or better)
* Algorithm::Munkres (version 0.07 or better)
* Algorithm::RandomMatrixGeneration (version 0.06 or better)
* Bit::Vector (version 6.3 or better)
* Math::SparseMatrix (version 0.02 or better)
* Math::SparseVector (version 0.04 or better)
* Set::Scalar (version 1.19 or better)
* Text::NSP (version 1.09 or better)
EXTERNAL PACKAGES (/External)
The following packages are not written in Perl and are developed outside
of the SenseClusters project. SVDPACKC is distributed as C source code,
and Cluto is distributed as pre-complied binaries for Linux and Solaris.
Please note that SVDPACK is optional - SenseClusters will run without it
(just don't use the --svd option). However, Cluto is mandatory,
SenseClusters will not be able to perform clustering without it.
* CLUTO (version 2.1.1 or better)
* SVDPACKC (Feb 2004 version or better, compiled with gcc 3.2.2, 3.2.3,
or 3.3.0)
MANUAL INSTALLATION OF CPAN MODULES
If the Bundle install does not succeed, you will need to install the
following modules manually:
* PDL (version 2.4.1 or better)
SenseClusters uses the Perl Data Language for efficient computations
and storage of high dimensional data structures.
It is available at: <http://search.cpan.org/dist/PDL/>
Note that if you have supervisor access on your machine, and have
the MCPAN Perl module available, you can install PDL automatically
via:
"perl -MCPAN -e 'install PDL';"
If you do not have supervisor access, you will need to install this
module locally. Note that you can configure the CPAN module to
install locally by setting PREFIX and LIB options to directories you
have read write authority over.
Note that PDL has quite a few dependencies, and can be tricky to
install. You may want to check with your system administrator and
see if they can install on your behalf before you tackle the local
install of PDL. All the other code mentioned here can be locally
installed quite routinely.
This is a good description of how to do local installs of Perl
modules: <http://www.perl.com/pub/a/2002/04/10/mod_perl.html>
* Bit::Vector (version 6.3 or better)
The Bit::Vector module is used with binary context vectors (via
--binary option in wrappers or program bitsimat.pl). This can be
downloaded from:
<http://search.cpan.org/dist/Bit-Vector/>
Note that the following installation instructions apply to all of the
CPAN modules, and will not be repeated in detail for each module.
If you have supervisor access, or have configured MCPAN for local
install, you can install via:
"perl -MCPAN -e 'install Bit::Vector';"
If not, you can, "manually" install by downloading the *.tar.gz
file, unpacking, and executing the following commands.
perl Makefile.PL PREFIX=/space/kulka020/Bit-Vector LIB=/space/kulka020/MyPerlLib
make
make test
make install
Note that the PREFIX and LIB settings are just examples to help you
create a local install, if you do not have supervisor (su) access.
You must include /space/kulka020/MyPerlLib in your PERL5LIB
environment variable to access this module when running.
* Text::NSP (version 1.09 or better)
SenseClusters uses the Ngram Statistics Package to select a variety
of lexical features.
Text::NSP is available at <http://search.cpan.org/dist/Text-NSP/>
"perl -MCPAN -e 'install Text::NSP';"
or manual installation.
* Set::Scalar (version 1.19 or better)
It is available at: <http://search.cpan.org/dist/Set-Scalar/>
"perl -MCPAN -e 'install Set::Scalar';"
or manual installation.
* Math::SparseVector (version 0.04 or better)
This is a Perl module that implements sparse vector operations.
It is available at: <http://search.cpan.org/dist/Math-SparseVector/>
"perl -MCPAN -e 'install Math::SparseVector';"
or manual installation.
* Math::SparseMatrix (version 0.02 or better)
This is a Perl module that implements sparse matrix operations, in
particular the sparse matrix transpose operation.
It is available at: <http://search.cpan.org/dist/Math-SparseMatrix/>
"perl -MCPAN -e 'install Math::SparseMatrix'";
or manual installation.
* Algorithm::Munkres (version 0.07 or better)
This is a Perl module that implements Munkres' solution to classical
Assignment Problem. This is used when carrying out evaluation of
discovered clusters with a provided gold standard.
It is available at: <http://search.cpan.org/dist/Algorithm-Munkres/>
"perl -MCPAN -e 'install Algorithm::Munkres';"
or manual installation.
* Algorithm::RandomMatrixGeneration (version 0.06 or better)
This is a Perl module that generates random matrix given the row and
column marginals. This is required for SenseClusters to run the
Adapted Gap Statistic in clusterstopping.pl.
It is available at:
<http://search.cpan.org/dist/Algorithm-RandomMatrixGeneration/>
"perl -MCPAN -e 'install Algorithm::RandomMatrixGeneration';"
or manual installation.
MANUAL INSTALLATION OF EXTERNAL PACKAGES
Please note that we provide a modified version of SVDPACK in
/External/SVDPACKC that makes all the changes described below. You
should be able to compile and install this code via the
External/install.sh script. If that fails you can follow the steps
described in the install script manually. If for some reason you
would prefer to start with a fresh copy of SVDPACKC, you can follow
the directions below (which also explain the changes we have made
and included in SenseClusters /External/SVDPACKC).
SVDPACKC (Feb 2004 version or better)
SVDPACKC is a C program that performs SVD. It is available for
download from <http://www.netlib.org/svdpack>. SVDPACKC does not
have a version number associated with it, but check the files in
your download to make sure they are dated from at least Feb 2004.
The version we include and modify in /External/SVDPACKC is the Feb
2004 version.
Please note that you should use version 3.2.2, 3.2.3, or 3.3.0 of
the gcc compiler. Segmentation faults results if you use version 4
or better. We are currently investigating the use of SVDLIBC as an
alternative for our SVD processing.
While installing SVDPACKC, you need to modify the following files
(already done in /External) :
1. In las2.c, uncomment the following line
/* #define UNIX_CREAT */
if you are running on a Unix or Linux platform.
2. In las2.h, modify the default values of constants LMTNW, NMAX and
NZMAX to some larger numbers such that -
NMAX = Maximum size of the feature space before reduction
(we set this to 30,000)
NZMAX = Maximum possible number of Non-zero entries
(we assume our 30,000 x 30,000 matrix is at most 1% dense
and hence NZMAX = 30,000 x 30,000 / 100 = 9,000,000)
LMTNW = Maximum Work Space Area required
= 6*NMAX + 4*NMAX + 1 + NMAX*NMAX
(we set LMTNW = 900300001 for a
1% dense 30,000 x 30,000 matrix)
3. Modify the file makefile so that ANSI C is used.
CC = gcc -ansi
[Please use gcc version 3.2.2, 3.2.3, or 3.3.0 when compiling SVDPACKC.
gcc versions 4.0.0 and above appear to result in segmentation faults.]
4. Run 'make las2' after the above modifications are done in las2.h,
las2.c, and makefile.
Testing SVDPACKC
The following steps will will help you check that SVDPACKC is
installed correctly.
# unzip the sample belladit.gz data file that comes with SVDPACKC
gunzip belladit.gz
# copy this as the input matrix to SVDPACKC
cp belladit matrix
# run las2 to test if everything is ok
las2
# this will not produce any output to STDOUT, but it should create 2
# output files - lav2 (binary) and lao2 (text)
CLUTO (version 2.1.1 or better)
The script External/install.sh will attempt to retrieve and install
Cluto automatically. If that fails, you can follow the steps outline
in the install script, or the instructions below.
SenseClusters uses CLUTO to support extensive clustering options,
analysis and visualization. CLUTO is freely available from
<http://www-users.cs.umn.edu/~karypis/cluto/>
If you run on both Linux and Solaris platforms, you will need to set
your path slightly differently each time to allow SenseClusters to
run. The following code in your .cshrc file will take care of this.
set OSNAME=`uname -s`
if ($OSNAME == "SunOS") then
set path = (PATH_2_CLUTO/Sun $path)
else if ($OSNAME == "Linux") then
set path = (PATH_2_CLUTO/Linux $path)
else echo "lost"
endif
where, PATH_2_CLUTO is a complete path to the directory where CLUTO
is downloaded and unpacked. If you only run on Solaris or Linux,
then of course you can just set the path with the appropriate
statement from above.
GCLUTO [optional]
Users interested in graphical visualization of clusters are
encouraged to try GCLUTO which is also freely down-loadable from
<http://www-users.cs.umn.edu/~karypis/cluto/gcluto/index.html>
To use GCLUTO, you will require the libglut.so.3 library installed
on your system. These can be downloaded from -
<http://at.rpmfind.net/opsys/linux/RPM/libglut.so.3.html>
CORE SENSECLUSTERS INSTALLATION
Note that SenseClusters can be installed via the Bundle command
described in the SYNOPSIS. If for some reason that fails, you can
proceed as follows:
To install the core of SenseClusters, if you have su or sudo access
(root user), then you can install via :
"perl -MCPAN -e 'install Text::SenseClusters';"
Or you can install manually as follows:
perl Makefile.PL
make
make install
cd Testing
csh ./testall.sh
cd ..
Note that the testall.sh scripts will not be run via automatic
installation. If you do not install manually, you should go back and
run the test scripts just to verify that everything is working as
expected.
The exact location where SenseClusters will be installed depends on
your system configuration. A message will be printed out after 'make
install' telling you exactly where it was installed.
Local Installation of Core SenseClusters programs
If you are not able to log in as su or sudo (to be the root user),
then you may need to install SenseClusters in a local directory that
you own and have permissions to read and write into. You can proceed
as above, except that you will need to provide PREFIX and LIB
options for your Makefile.PL command, as in:
"perl Makefile.PL PREFIX = /YOUR/DIR LIB=/YOUR/DIR/lib"
This will set up a Makefile that will install the core SenseClusters
programs (*.pl) into "/YOUR/DIR/bin/". You will need to set your
path to specifically include this directory.
The Sensecluster.pm module will be installed into "/YOUR/DIR/lib".
You will need to set your PERL5LIB environment variable to have this
directory included in your @INC array (which defines the directories
that Perl searches for modules).
You man pages will be installed in directories like
"/YOUR/DIR/share/man/" (Linux) or "/YOUR/DIR/man/" (Solaris). You
will need to set your MANPATH to include these.
Note that the exact locations will be shown after executing 'make
install' command. Please double check the recommended settings for
PATH and MANPATH there as those will be tailored to your system.
C SHELL (csh) SETUP
If you install without root or superuser access, you will need to
set the paths of the dependent packages mentioned previously. The
following is an example of how you might set your paths before using
SenseClusters if you are using the C shell (csh). If you use another
shell then you will need to modify this accordingly.
This assumes that Perl and PDL have been installed by your system
administrator and you do not need to set paths to find them. In
general we would recommend that Perl and PDL be installed with root
access as it's more simple that way.
Assume that all of the external C packages (SVDPACKC, Cluto) have
been installed in directories beneath /space/kulka020 (our home
directory for this example). It also assumes that all of the CPAN
modules have been installed in /space/kulka020/MyPerlLib. In other
words, it is assumed that all CPAN modules were installed via the
following command:
"perl Makefile.PL PREFIX=/space/kulka020/Text-SenseClusters-1.00
LIB=/space/kulka020/MyPerlLib"
#######################################################################
# insert the following into ~/.cshrc and modify HOMEDIR and LIBHOME
#######################################################################
# local directory where we are installing everything
setenv HOMEDIR /space/kulka020
# library name extension used by Perl on our system
setenv LIBDIR /space/kulka020/MyPerlLib
# UMD developed code, we need to set their /bin directories in the PATH
setenv SENSECLUSTERS $HOMEDIR/Text-SenseClusters-1.00
setenv NSP $HOMEDIR/Text-NSP-1.09
# externally developed C code, directories contain executable code so must
# be included in PATH
setenv SVDPACK $HOMEDIR/SVDPACKC
setenv CLUTO $HOMEDIR/cluto-2.1.1
# pick the right version of Cluto (Solaris or Linux)
set OSNAME=`uname -s`
if ($OSNAME == "SunOS") then
setenv MYCLUTO $CLUTO/Sun
else if ($OSNAME == "Linux") then
setenv MYCLUTO $CLUTO/Linux
else echo "lost"
endif
# set the path that Perl searches for CPAN modules
setenv PERL5LIB $LIBDIR
# set the path that is searched for executables
set AKPATH = ($SVDPACK $NSP/bin $MYCLUTO $SENSECLUSTERS/bin .)
set path = ($AKPATH $path)
INSTALLING SenseClusters' Web-interface:
If you would like to setup the SenseClusters' web-interface locally
please refer to README.Web.pod for installation instructions.
SEE ALSO
The SenseClusters web page provides links to downloads, the web
interface, documentation, and CVS directories:
L<http://senseclusters.sourceforge.net>
We have three mailing lists available for SenseClusters:
* senseclusters-news provides announcements of new versions:
<http://lists.sourceforge.net/lists/listinfo/senseclusters-news>
* senseclusters-users allows users to post questions or bug
reports:
<http://lists.sourceforge.net/lists/listinfo/senseclusters-users
>
* senseclusters-developers is for implementation related
questions:
<http://lists.sourceforge.net/lists/listinfo/senseclusters-devel
opers>
AUTHORS
Ted Pedersen, University of Minnesota, Duluth
tpederse at d.umn.edu
Anagha Kulkarni, Carnegie Mellon University
Amruta Purandare, University of Pittsburgh
This document last modified by : $Id: INSTALL.pod,v 1.15 2008/03/31
16:19:57 tpederse Exp $
COPYRIGHT
Copyright (c) 2004-2008, Ted Pedersen
Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License,
Version 1.2 or any later version published by the Free Software
Foundation; with no Invariant Sections, no Front-Cover Texts, and no
Back-Cover Texts.
Note: a copy of the GNU Free Documentation License is available on
the web at <http://www.gnu.org/copyleft/fdl.html> and is included in
this distribution as FDL.txt.