From Documentation
Revision as of 14:31, 4 April 2012 by Isaac (Talk | contribs) (Created page with "{{Software |package_name=OCTAVE |package_description=Mostly compatible language with Matlab primarily intended for numerical computations |package_idnumber=28 }} <font color=red...")

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search
OCTAVE
Description: Mostly compatible language with Matlab primarily intended for numerical computations
SHARCNET Package information: see OCTAVE software page in web portal
Full list of SHARCNET supported software



NOTICE : On December 1st 2011 octave/3.4.0 will be removed from all sharcnet clusters since version octave/3.4.3 is now installed and contains significant performance enhancements. Also on Dec 1 the octave/3.2.4 module will no longer load by default on goblin and hound clusters.

For best performance load the latest version of octave by doing:

module unload octave
module load octave

On SHARCNET clusters Octave should only be run via the queuing system. Octave serial jobs bas best submitted to the serial queue on kraken assuming they use less than 1gb by doing:

sqsub -t -r 60m -f wha -o ofile.%J octave mycode.m

As of version 3.4.3 the sharcnet octave installation supports multithreading which based on initial testing can result in an order of magnitude speedup for octave codes that are specifically lapack and blas intensive (compared to simply using a single core in the serial queue) without making any changes to your code. Once the optimal number of processors is determined by scaling tests submit the job to the threaded queue for example:

sqsub -t -r 60m -n 16 -q threaded --mpp=1G -o ofile.%J time octave mycode.m

WORKING WITH ADDITIONAL PACKAGES

The "Octave Forge"[1] project provides extra packages for use with octave that can be downloaded and manually installed into your own sharcnet account. A list of packages which can be downloaded can be found "here"[2]. Consider setting up a directory called my_octave_sources in your sharcnet home account for all your downloaded packages as is assumed in the following example:

[jole@bul125:~] octave

octave:1>  help pkg

To install a package such as linear-algebra do

octave:2>  cd ~/my_octave_sources

octave:3>  pkg install linear-algebra-2.1.0.tar.gz

To then remove the linear-algebra do

 octave:4>  pkg uninstall linear-algebra

To list all installed packages do

 octave:5>  pkg list

To add a named packages to your path do

 octave:6>  pkg load name

To remove a named package from your path do

 octave:8>   pkg unload

To list all functions provided by a package do

 octave:9>  pkg describe -verbose all

To list functions provide by extra odepkg package do

 octave:10>  pkg describe odepkg

To list functions details for extra financial package

 octave:11>  pkg describe financial -verbose

To get documentation for a topic such as sin do

 octave:12>   doc sin

To get help using the doc command do

octave:13>   help doc

To get documentation for a topic such as matrix do

 octave:14>  doc matrix

To get documentation for any topic run the doc command alone

octave:15>  doc

READING OR WRITING FILES

There are two strategies for handling file "input and output"[3] described in the "Octave manual"[4]. The following are some demonstrating the "simple file I/O"[5] approach

 
save myiofile.dat A B C
save ("-text", "myiofile.dat", "A", "B", "C")
save ("-binary", "myiofile.dat", "A", "B", "C")
load myiofile.dat
load ("-text", "myiofile.dat", "A", "B", "C")
load ("-binary", "myiofile.dat", "A", "B", "C")

where A, B and C are a potential mix of entities such as scalars, vectors or matrices. Note that for large files the binary format is strongly recommended to both minimize disk space and file read/write wallclock times.

A COMPLETE PACKAGE EXAMPLE

This example assumes you first create two directories in your home account one being named ~/my_octave_packages and the other ~/my_octave_sources then download the required three tar.gz files into the latter from here [6]

[roberpj@orc-login1:~] module unload octave
[roberpj@orc-login1:~] module load octave
[roberpj@orc-login1:~] octave
GNU Octave, version 3.4.3
octave:1> pkg prefix ~/my_octave_packages
ans = /home/roberpj/my_octave_packages
octave:2> cd my_octave_sources
octave:3> ls
miscellaneous-1.0.11.tar.gz  optim-1.0.17.tar.gz  struct-1.0.9.tar.gz
octave:4> pkg install miscellaneous-1.0.11.tar.gz optim-1.0.17.tar.gz struct-1.0.9.tar.gz
octave:5> pkg list
Package Name   | Version | Installation directory
---------------|---------|-----------------------
miscellaneous *|  1.0.11 | /home/roberpj/my_octave_packages/miscellaneous-1.0.11
        optim *|  1.0.17 | /home/roberpj/my_octave_packages/optim-1.0.17
       struct *|   1.0.9 | /home/roberpj/my_octave_packages/struct-1.0.9


MATLAB COMPATIBILITY

The online wiki resources "Octave Wiki"[7] or "Wikibook"[8] provide good introductory explanations of code compatibility between Octave and Matlab.

LARGE MEMORY ARRAY ALLOCATION TESTING

The following command sequence should successfully run on hound or tope. Since tope is recommended for interactive use and has 32gb of memory the test will be demonstrated there:

[mypc ~] ssh tope.sharcnet.ca

[tope ~] module load octave/3.4.0

[tope ~] octave
GNU Octave, version 3.4.0
octave:3> N=4e8; A=[1:N]; B=[2*(1:N)]; C=A+B; C(1); C(N); 3*N 
ans =  1.2000e+09

[tope ~] top
24985 roberpj   24   0 9406m 9.0g  12m S  0.0 32.6   0:34.34 octave