2012HackSessionNotes¶

SMALLbox Add-ons¶

Installing an add on¶

This section refers to the incoherentdl algorithm code tagged as ver_1.1 in the project’s Mercurial repository. To download this version please go to the project’s homepage in http://code.soundsoftware.ac.uk/projects/incoherentdl.
The incoherent dictionary learning add-on includes algorithms for learning dictionaries that are both adapted to the training data and that exhibit a low mutual coherence (defined as the maximum absolute inner product between any two different atoms) [27]. It extends
26
SMALLbox v2.0 User Documentation 27
the functionalities of the SMALL_two_step_DL toolbox which is included in the SMALLBox distribution, and is an example showing the capabilities of SMALLbox add–ons.
These are the steps needed to interface SMALLbox add-ons to the core SMALLbox dis- tribution:

• Download the add-on distribution from: http://code.soundsoftware.ac.uk/projects/incoherentdl/files
  • The version you wish to download is version 1.1
• Place the add-on code in any convenient location of your file system, and add this to your MATLAB path
  • addpath(genpath('/Users/luisf/incoherentdl-v1_1'))
• Modify the relevant files in the folder {SMALLBox Root}/config/
  • modify/create SMBOX/config/SMALL_learn_config_local.m
    • cp SMBOX/config/SMALL_learn_local.m SMBOX/config/SMALL_learn_config_local.m
    • add the following code to the local config file:

elseif strcmpi(DL.toolbox , 'SMALL_incoherentDL')
  DL=SMALL_incoherentDL ( Problem , DL ) ;
   % we need to make sure that columns are normalised to % unit lenght .
  for i = 1: size(DL.D,2)
    DL.D(: ,i)=DL.D(: ,i)/norm(DL.D(: ,i));
  end
  D = DL.D;

• Run the SMALLboxInit.m script to set environmental variables.
• Run any function or script contained in the add-on.
  • example:
    • load test_mocod.mat (incoherentdl-v1_1/results/test_mocod.mat)
    • open SMALL_test_mocod.m
    • run the code from the block that starts on line 110 (with the comment Plot Results)
      • this will display two coherence/signal to noise ratio plots based on results previously computed

Creating a new add on¶