diff core/tools/machine_learning/get_itml_deltas.m @ 0:e9a9cd732c1e tip

first hg version after svn

--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/core/tools/machine_learning/get_itml_deltas.m	Tue Feb 10 15:05:51 2015 +0000
@@ -0,0 +1,85 @@
+function [X, C, idx] = get_itml_deltas(r, in)
+% [X, C, idx] = get_itml_deltas(r, in)
+
+%ITML Specs
+% C: 4 column matrix
+%      column 1, 2: index of constrained points.  Indexes between 1 and n
+%      column 3: 1 if points are similar, -1 if dissimilar
+%      column 4: right-hand side (lower or upper bound, depending on 
+%                   whether points are similar or dissimilar)
+%
+% X: (n x m) data matrix - each row corresponds to a single instance
+% ---
+% NOTE: X is thus input in transposed shape for the ITML algorithm
+% ---
+
+% ---
+% NOTE: this preallocation is not complete
+% ---
+X = zeros(size(in,1), 0);
+C = zeros(0,4);
+idx = zeros(0,2);
+
+for i = 1:size(r,1)
+    
+    % feature indexing
+    a = i;
+    
+    % check if ranking is valid
+    if ~isempty(r{i,1}) && ~isempty(r{i,2})&& ...
+        isempty(intersect(r{i,1}, r{i,2}));
+    
+        % ---
+        % NOTE / TODO: the follwing is intended for compability
+        %  both sides of the ranking may have more than one entry.
+        %  for the MTT database, the ranking may be correct, but the 
+        %  inequalities build from non-singular rankings are not
+        %  based on the actual data
+        % ---
+        for j = 1:numel(r{i,1})
+            b = r{i,1}(j);
+            
+            for k = 1:numel(r{i,2})
+                c = r{i,2}(k);
+
+                % ---
+                % get vector deltas
+                % ---
+                [dab] = in(:,a) - in(:,b);
+                [dac] = in(:,a) - in(:,c);
+                
+                % ---
+                % save deltas in new feature matrix
+                % TODO: this method has duplicate entries
+                %  if the pairs appear more than once
+                %  index the data set and use more efficiently!!!
+                % ---
+                X = [X dab];
+                idx(end+1,:) = [a b];
+                iab = size(idx, 1);
+                
+                X = [X dac];
+                idx(end+1,:) = [a c];
+                iac = size(idx, 1);
+                
+                % ---
+                % NOTE:
+                % in terms of the constraint,
+                %  this should mean: dac - dab >= 1
+                %
+                % 4th position cannot be 0, converges to Inf if > 1
+                % -1,-1 learns the opposite of what constraitns say
+                % ---
+                C(end+1, :) = [iab iac -1 -1];
+            end
+        end
+    end
+end
+
+% % ---
+% % NOTE: here, we transpose the X for usage i nthe training
+% % ---
+% X = X';
+
+end
+