find_specificity

 ==============================================================
 FIND_SPECIFICITY: Find the specificity of a model.

 Code written by Katherine Smith, 2003

    GENERAL

      [mean_specificity, std_specificity, all_specificities] =
       find_specificity(model, training_data, n_iterations, 
       ref_points_vec, varargin)

    INPUT/S

      -model:
           The model for which specificity is evaluated.

      -training_data:
           The training data used to evalaute the specificity
           of that model.

      -n_iterations:
           ?The number of iterations used in the evaluation process?

      -ref_points_vec:
           ?Reference points?

      -varargin:
           ?Variance argument input?
           
    OUTPUT/S

      -mean_specificity:
           The mean specificity.

      -std_specificity:
           The standard deviation of specificities.

      -all_specificities:
           All the specificity values found.

    PENDING WORK

      -Resolve meaning of inputs and outputs.

    KNOWN BUG/S

      -None.

    COMMENT/S

      -Smith: Construct examples from model and measure 
       the distance from the nearest example in training set
      -Full definition given by Rh. Davies.

    RELATED FUNCTION/S

      FIND_GENERALISABILITY

    ABOUT

      -Created:     November 23rd, 2003
      -Last update: Novermber 25th, 2003
      -Revision:    0.0.2
      -Author:      R. S. Schestowitz, University of Manchester
 ==============================================================

0001 function [mean_specificity, std_specificity, all_specificities] = find_specificity(model, training_data, n_iterations, ref_points_vec, varargin)
0002 % ==============================================================
0003 % FIND_SPECIFICITY: Find the specificity of a model.
0004 %
0005 % Code written by Katherine Smith, 2003
0006 %
0007 %    GENERAL
0008 %
0009 %      [mean_specificity, std_specificity, all_specificities] =
0010 %       find_specificity(model, training_data, n_iterations,
0011 %       ref_points_vec, varargin)
0012 %
0013 %    INPUT/S
0014 %
0015 %      -model:
0016 %           The model for which specificity is evaluated.
0017 %
0018 %      -training_data:
0019 %           The training data used to evalaute the specificity
0020 %           of that model.
0021 %
0022 %      -n_iterations:
0023 %           ?The number of iterations used in the evaluation process?
0024 %
0025 %      -ref_points_vec:
0026 %           ?Reference points?
0027 %
0028 %      -varargin:
0029 %           ?Variance argument input?
0030 %
0031 %    OUTPUT/S
0032 %
0033 %      -mean_specificity:
0034 %           The mean specificity.
0035 %
0036 %      -std_specificity:
0037 %           The standard deviation of specificities.
0038 %
0039 %      -all_specificities:
0040 %           All the specificity values found.
0041 %
0042 %    PENDING WORK
0043 %
0044 %      -Resolve meaning of inputs and outputs.
0045 %
0046 %    KNOWN BUG/S
0047 %
0048 %      -None.
0049 %
0050 %    COMMENT/S
0051 %
0052 %      -Smith: Construct examples from model and measure
0053 %       the distance from the nearest example in training set
0054 %      -Full definition given by Rh. Davies.
0055 %
0056 %    RELATED FUNCTION/S
0057 %
0058 %      FIND_GENERALISABILITY
0059 %
0060 %    ABOUT
0061 %
0062 %      -Created:     November 23rd, 2003
0063 %      -Last update: Novermber 25th, 2003
0064 %      -Revision:    0.0.2
0065 %      -Author:      R. S. Schestowitz, University of Manchester
0066 % ==============================================================
0067 
0068 args = u_packargs(varargin, 0,'n_modes', size(model.params,2));
0069 
0070 best_yet = ones(n_iterations,1)*size(training_data,1); %construct vector of as many ones as iternations and multiply these ones by the size of the training data
0071 stds = sqrt(model.variances); % get square roots of model variances
0072 stds = real(stds); % convert these to real numbers
0073 for i=1:n_iterations % for input number of iterations
0074   params = zeros(1,size(model.params,2)); 
0075          % paramters set to 0, their size being the length of second bunch of model parameters
0076   invalid_params = 1; % set initially to invalid
0077   while(invalid_params) % and iterate while no validity proven
0078     params(1:args.n_modes) = randn(1,args.n_modes).*stds(1:args.n_modes);
0079       % for all modes, set them to a number in the range of their total number * the variance in the position of number of modes
0080     [example,points] = construct_model_example(params, model, ref_points_vec);
0081      % create a few examples given the parameters
0082     if(~is_invalid(points)) % if all our valid
0083       invalid_params = 0;  % break out of loop
0084     end
0085   end
0086 
0087   for j=1:size(training_data,2)
0088     distance = msd(example, training_data(:,j)); % get mean squred distance
0089     if(distance < best_yet(i)) % and try to find the closest yet
0090       best_yet(i) = distance; 
0091     end
0092   end
0093   if(best_yet(i) > 10)
0094     disp('stop here');
0095   end
0096 end
0097 mean_specificity = mean(best_yet);
0098 std_specificity = std(best_yet);
0099 all_specificities = best_yet;