Description of slpweval

Home > sltoolbox > core > slpweval.m

slpweval

PURPOSE

SLPWEVAL Perform pairwise computation

SYNOPSIS

function M = slpweval(X1, X2, f, varargin)

DESCRIPTION

SLPWEVAL Perform pairwise computation

 $ Syntax $
   - M = slpweval(X1, X2, f)
   - M = slpweval(X1, X2, f, ...)

 $ Arguments $
   - X1:       the matrix of vectors serving as first argument of f
   - X2:       the matrix of vectors serving as second argument of f
   - f:        the function maps two vectors to a single scalar value
   - M:        the matrix of pairwise evaluaton results

 $ Description $
   - M = slpweval(X1, X2, f) takes vector arguments from X1 and X2, and 
     computes the pairwise evaluation result with f. The vectors in X1
     and X2 are stored in a column-wise manner. Suppose X1 and X2 have 
     m and n columns respectively, then the resultant matrix M would be
     of size m x n, with the M(i, j) = f(X1(:,i), X2(:,j)). 
   
   - M = slpweval(X1, X2, f, ...) conducts the computation with extra
     parameters to f, i.e. M(i, j) = f(X1(:,i), X2(:,j), ...).

 $ Remarks $
   - The vector length of the vectors in X1 and X2 are not necessarily
     equal. The requirment on their dimensions depends on the callback
     function f.
   - For efficiency, the function would invoke f to evaluate in batch.
     Thus f should support batch-evaluation. When the input arguments
     to f have n columns, f should return an 1 x n row vector.

 $ History $
   - Created by Dahua Lin on Apr 21, 2006
   - Modified by Dahua Lin on Sep 10, 2006
       - make some minor changes to suppress warnings

CROSS-REFERENCE INFORMATION

This function calls:

• raise_lackinput RAISE_LACKINPUT Raises an error indicating lack of input argument
• slignorevars SLIGNOREVARS Ignores the input variables

This function is called by:

• slkernel SLKERNEL Computes the kernel for samples

SOURCE CODE

0001 function M = slpweval(X1, X2, f, varargin)
0002 %SLPWEVAL Perform pairwise computation
0003 %
0004 % $ Syntax $
0005 %   - M = slpweval(X1, X2, f)
0006 %   - M = slpweval(X1, X2, f, ...)
0007 %
0008 % $ Arguments $
0009 %   - X1:       the matrix of vectors serving as first argument of f
0010 %   - X2:       the matrix of vectors serving as second argument of f
0011 %   - f:        the function maps two vectors to a single scalar value
0012 %   - M:        the matrix of pairwise evaluaton results
0013 %
0014 % $ Description $
0015 %   - M = slpweval(X1, X2, f) takes vector arguments from X1 and X2, and
0016 %     computes the pairwise evaluation result with f. The vectors in X1
0017 %     and X2 are stored in a column-wise manner. Suppose X1 and X2 have
0018 %     m and n columns respectively, then the resultant matrix M would be
0019 %     of size m x n, with the M(i, j) = f(X1(:,i), X2(:,j)).
0020 %
0021 %   - M = slpweval(X1, X2, f, ...) conducts the computation with extra
0022 %     parameters to f, i.e. M(i, j) = f(X1(:,i), X2(:,j), ...).
0023 %
0024 % $ Remarks $
0025 %   - The vector length of the vectors in X1 and X2 are not necessarily
0026 %     equal. The requirment on their dimensions depends on the callback
0027 %     function f.
0028 %   - For efficiency, the function would invoke f to evaluate in batch.
0029 %     Thus f should support batch-evaluation. When the input arguments
0030 %     to f have n columns, f should return an 1 x n row vector.
0031 %
0032 % $ History $
0033 %   - Created by Dahua Lin on Apr 21, 2006
0034 %   - Modified by Dahua Lin on Sep 10, 2006
0035 %       - make some minor changes to suppress warnings
0036 %
0037 
0038 %% parse and verify input arguments
0039 
0040 if nargin < 3
0041     raise_lackinput('slpweval', 3);
0042 end
0043 [d1, n1] = size(X1);
0044 [d2, n2] = size(X2);
0045 slignorevars(d1, d2);
0046 
0047 %% compute
0048 
0049 % prepare output matrix
0050 M = zeros(n1, n2);
0051 
0052 if n1 > n2      % expand each column in X2 to n1 copies
0053  
0054     inds_e = ones(1, n1);
0055     for i = 1 : n2        
0056         x2 = X2(:, i);
0057         X2e = x2(:, inds_e);        
0058         M(:, i) = feval(f, X1, X2e, varargin{:})';            
0059     end
0060     
0061 else            % expand each column in X1 to n2 copies
0062     
0063     inds_e = ones(1, n2);
0064     for i = 1 : n1
0065         x1 = X1(:, i);
0066         X1e = x1(:, inds_e);
0067         M(i, :) = feval(f, X1e, X2, varargin{:});
0068     end    
0069     
0070 end
0071

---

Generated on Wed 20-Sep-2006 12:43:11 by m2html © 2003