Skip to content

Latest commit

 

History

History
99 lines (56 loc) · 6.54 KB

installation.mediawiki

File metadata and controls

99 lines (56 loc) · 6.54 KB
Raw

Installation instructions for pmtk3

Table of Contents

Installation

Download the latest zip file from github, or clone the repository.

Start Matlab/Octave. Suppose the directory where you unzipped pmtk is called 'C:\pmtk3'. Then type the following (you will need to modify the directory names if you unzipped these files to a different place). 

  cd C:\pmtk3
  initPmtk3

You should see something like this. 

  initializing pmtk3
  downloading 38 packages to pmtk3/pmtksupportCopy from pmtksupport.googlecode.com - this may take a few      minutes 
  downloading GGM-GWishart............success
  ...

If this fails for some reason, you may need to manually create the following directory 

  C:\pmtk3\pmtksupportCopy

Then try re-running initPmtk3 again (from within matlab).

To test the system, type 

  testPmtk3

This command will check the packages are installed correctly, automatically downloads a few datasets, and then runs a few demos; this takes about a minute.

If you have more patience, you can try this: 

  runDemos

This takes about 1 hour. The output produced by running all these demos is shown here.

Note: Pmtk3 has only been tested on Matlab 2009a or newer. It is possible it will run on 2008a or older versions (see matlab release history), but this is not guaranteed. It has been tested on windows 7, mac os X, and ubuntu linux. Octave compatibility is discussed at the end of this page.

PMTK3 relies on various support packages, which are automatically downloaded from pmtksupport by initPmtk3.m if they are not already on your path. Once downloaded, the files are stored locally in the 'pmtksupportCopy' directory and added to your path, so the next time the local copy is used. Unlike the data sets, the support packages are downloaded all at once, not on demand. However, you can check them all out at once if you prefer. In this case, you should edit the config-local.txt file to specify the location of the support directory.

Mex files

Most of PMTK is written in Matlab, so should run on any system. However, a few functions also have C versions, or only have C versions, for speed. We have included precompiled versions of these files. See this page for details on recompiling them, if necessary.

Libdai interface

PMTK3 includes an optional interface to the libdai graphical model inference package, written in C++ by Joris Mooij. We have compiled windows (32bit and 64bit), and (ubuntu) linux 32 bit versions of libdai, which are included in pmtkSupport, and hence automatically available. Users of other systems will need to compile libdai manually. You can find libdai's installation instructions here, and our notes on compiling libdai for windows here. Once compiled, add the *.mex* files from libdai's matlab subfolder to the pmtkSupport libdai directory. If you generate mex files for other OS's, please send them to us so that we can include them for others to use.

Graphviz

The graphViz4Matlab package, which is included in pmtkSupport, can be used to visualize (and edit) network structures. For best results, you should install the latest (free) version of Graphviz. This feature requires you to have Matlab 2008a or newer. It is not necessary to have graphviz to use pmtk, but it is useful if you want to visualize the results of graphical model structure learning, and/or run some of the demos. After installation, modify PMTKgvizPath in your config.txt file, and/or add this directory to your path using addtosystempath.m. On windows systems the file is usually stored at `C:\program files\Graphviz2.26\bin`. Ubuntu/Debian users can install graphviz by typing `sudo apt-get install graphviz` at the command prompt.

Octave issues

PMTK is only about 75% octave compatible. If you run into problems in Octave, check out here or here for mex files,as well as the points below.

Octave does not always seem to include the current working directory on the path. You may need to type e.g. `addpath('C:\pmtk3')`, before typing `initPmtk3`. It also doesn't seem to like spaces in file paths.

Octave users on windows systems need to have an unzip.exe file on the system path. If you don't, you can download one from info-zip. Whatever unzip program you use, it must be called "unzip" and be on your _system_ path.

Octave under linux can sometimes fail to include all the of the necessary directories after the call to initPmtk3. If this happens you may need to manually type `restoredefaultpath(); addpath(genpath(pwd))` after running initPmtk3() for the first time, and then `addpath(genpath(pwd))` subsequent times instead of `initPmtk3`. Note that under Octave, `genpathPMTK` is very slow and does not work reliably, use `genpath` instead.

Warning messages produced by glmnet

When running runDemos, you may get various warning messages produced by the glmnet package. You can safely ignore them. Some of the more common ones are listed below, FYI.

You may get this warning message 

  Warning: In the directory "C:\kmurphy\GoogleCode\pmtkSupport\glmnet-matlab", glmnetMex.mexw32 now shadows glmnetMex.dll.

You can safely ignore this. (It just means there are two versions of the executable on the path: .mexw32 is the 32 bit Windows version for new versions of Matlab, and .dll is for older versions of Matlab.)

You may get this warning message 

   cannot run lassoPathProstateClassif without glmnet; skipping

This just means you don't have a glmnet binary on your machine.

You may get this message or something similar: 

   Warning: from glmnet fortran code - Convergence for 33th lambda value not reached after maxit=100
   iterations; solutions for larger lambdas returned

Again, you can safely ignore this. (It is caused by the glmnet package when it attempts to compute the full regularization path. This functionality is not central to anything used by pmtk.)