Merge branch 'master' of ssh://github.com/covartech/prt

.gitignore

@@ -1 +1,2 @@
-.DS_Store
+.DS_Store
+*.JPG

engine/prtAlgorithm.m

@@ -515,6 +515,25 @@ function plotAsClassifier(self)
                 error('prt:prtAlgorithm:plotAsClassifier','This prtAlgorithm cannot be plotted as a classifier');
             end
         end
+        function plotAsRegressor(self)
+            % plotAsRegressor(self)
+            %   Plot an algorithm as though it were a regressor - e.g.,
+            %   build the decision surface and visualize it.  Valid for
+            %   algorithms trained with data sets with 3 or fewer features,
+            %   and when the *very last* action is a prtRegress.
+            %
+            % e.g.
+            %   ds = prtDataGenBimodal;
+            %   algoKmeans = prtPreProcKmeans('nClusters',4) + prtClassLogisticDiscriminant;
+            %   algoKmeans = train(algoKmeans,ds);
+            %   algoKmeans.plotAsClassifier;
+
+            if isPlottableAsRegressor(self)
+                plot(prtUtilRegressAlgorithmWrapper('trainedAlgorithm',self));
+            else
+                error('prt:prtAlgorithm:plotAsRegressor','This prtAlgorithm cannot be plotted as a regressor');
+            end
+        end
         function tf = isPlottableAsClassifier(self)
 
             tf = false;
@@ -533,6 +552,24 @@ function plotAsClassifier(self)
                 end
             end
         end
+        function tf = isPlottableAsRegressor(self)
+
+            tf = false;
+            if isempty(self.dataSetSummary)
+                return
+            end
+
+            if self.dataSetSummary.nFeatures <= 3
+                if sum(self.outputNodes)==1
+                    lastNodes = self.connectivityMatrix(find(self.outputNodes,1,'first'),:);
+                    if sum(lastNodes)==1
+                        if isa(self.actionCell{find(lastNodes,1,'first')-1},'prtRegress')
+                            tf = true;
+                        end
+                    end
+                end
+            end
+        end
 
 
         function [optimizedAlgorithm,performance] = optimize(Obj,DataSet,objFn,tag,parameterName,parameterValues)

plot/util/prtPlotUtilAlgorithmGui.m

@@ -123,6 +123,8 @@ function patchButtonDownFunction(hObject, eventData, blockIndex) %#ok<INUSL>
                     plot(BlockObject);
                 elseif (blockIndex == length(Layout.Blocks)) && algo.isPlottableAsClassifier
                     plotAsClassifier(algo);
+                elseif (blockIndex == length(Layout.Blocks)) && algo.isPlottableAsRegressor
+                    plotAsRegressor(algo);
                 end
 
             otherwise % case 'normal'
@@ -192,7 +194,7 @@ function placeBlockFunction(hObject, eventData, BlockObject, position, iBlock, b
             set(Layout.Blocks(iBlock).handle,'lineWidth',4);
         end
 
-        if (iBlock == Layout.nBlocks) && algo.isPlottableAsClassifier
+        if (iBlock == Layout.nBlocks) && (algo.isPlottableAsClassifier || algo.isPlottableAsRegressor)
             set(Layout.Blocks(iBlock).handle,'lineWidth',4);
         end
 

regress/prtRegress.m

@@ -41,13 +41,27 @@
 
 
             assert(Obj.isTrained,'Regressor must be trained before it can be plotted.');
-            assert(Obj.dataSetSummary.nFeatures < 2, 'nFeatures in the training dataset must be 1');
 
             [OutputDataSet, linGrid] = runRegressorOnGrid(Obj);
 
             colors = Obj.plotOptions.colorsFunction(Obj.dataSetSummary.nTargetDimensions);
             lineWidth = Obj.plotOptions.lineWidth;
-            HandleStructure.regressorPlotHandle = plot(linGrid,OutputDataSet.getObservations,'color',colors(1,:),'lineWidth',lineWidth);
+
+            switch Obj.dataSetSummary.nFeatures
+                case 1
+                    HandleStructure.regressorPlotHandle = plot(linGrid,OutputDataSet.getObservations,'color',colors(1,:),'lineWidth',lineWidth);
+                case 2
+                    sz = ones(1,Obj.dataSetSummary.nFeatures)*Obj.plotOptions.nSamplesPerDim(Obj.dataSetSummary.nFeatures);
+                    HandleStructure.regressorPlotHandle = surf(...
+                        reshape(linGrid(:,1),sz),...
+                        reshape(linGrid(:,2),sz),...
+                        reshape(OutputDataSet.getObservations,sz));
+                    HandleStructure.regressorPlotHandle.EdgeColor = colors(1,:);
+                    HandleStructure.regressorPlotHandle.FaceColor = 'none';
+                    HandleStructure.regressorPlotHandle.LineWidth = lineWidth;
+                otherwise
+                    error('nFeatures in the training dataset must be 1 or 2');
+            end
 
             holdState = get(gca,'nextPlot');
             if ~isempty(Obj.dataSet)

regress/prtRegressBayesianLinear.m

@@ -15,12 +15,11 @@
         includeBias = true;
 
         priorWeightsMean = 0; % If scalar will be initialized by replication
-        priorWeightsPrecision = eps; % If scalar will be 
+        priorWeightsPrecision = eps; % If scalar will be initialized by replication
 
         priorPrecisionA = 0.1;
         priorPrecisionB = 0.1;
 
-
         mu
         lambda
         a
@@ -36,10 +35,11 @@
 
     methods
 
-          % Allow for string, value pairs
         function self = prtRegressBayesianLinear(varargin)
+            % Allow for string, value pairs
             self = prtUtilAssignStringValuePairs(self,varargin{:});
         end
+
     end
 
     methods (Access = protected, Hidden = true)
@@ -65,9 +65,7 @@
            end
            y = ds.Y;
 
-           XtX = X'*X;
-
-           self.lambda = (XtX + lambda0);
+           self.lambda = (X'*X + lambda0);
            self.mu = (mu0*lambda0+ y'*X)/(self.lambda);
 
            self.a = a0 + size(X,1)/2;

regress/prtRegressGp.m

@@ -119,6 +119,7 @@
             if ~isempty(self.meanRegressor)
                 dataSetEst = self.meanRegressor.run(dataSet);
             else
+<<<<<<< Updated upstream
                 % The mean regressor will output dataSet.X of the same size
                 % as dataSet.Y, with all zeros!  dataSetEst.Y should be
                 % dataSet.Y.  Note - at test-time, there may not be
@@ -127,6 +128,9 @@
                 % (size(dataSet.Y,2)) is 1, even if there werent any
                 % targets provided (dataSet.Y is empty)
                 dataSetEst = prtDataSetRegress(zeros(size(dataSet.X,1),1),dataSet.Y);
+=======
+                dataSetEst = prtDataSetRegress(zeros(dataSet.nObservations,1),zeros(dataSet.nObservations,1));
+>>>>>>> Stashed changes
             end
             dataSetTargetResiduals = dataSet;
             if nargout > 1