diff code/treeLinkFeatures.m @ 32:4bdcab1e821c

tidy up directory

--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/code/treeLinkFeatures.m	Wed Mar 15 11:33:55 2017 +0000
@@ -0,0 +1,58 @@
+function [linkList, featureList]= treeLinkFeatures(data, depthThresh, featureNames)
+%% [linkList, featureList]= treeLinkFeatures(data, depthThresh, featureNames)
+% given a dataset, a hierarchical cluster of the data is produced, and then
+% the data is traversed, such that, for each split in the data, a set of
+% features are produced, which are the ranked features that can be used to
+% separate the given dataset at that point.
+% data is the nxm matrix of content, n is the number of samples and m is
+% the number of features.
+% depthThresh is a list of the range of tree depths to traverse from the
+% aglomerative clustering tree. A single value of depthThresh, will assume
+% 1:depthThresh. For analysis of a single layer of the tree, pass a list of
+% two values, both of which are the layer to be analysed.
+% feature names is the list of features, so that grown trees have suitable
+% names. No feature names will result in the feature number being returned.
+% featureList corresponds to the rows in linkList, with the form column 1
+% is the 5 most relevant features, column 2 is the depth and column 3 is a
+% decision classification tree for the decision - perhaps this should be in
+% the form of a struct instead?
+
+
+
+if(nargin < 3)
+    featureNames = 1:size(data,2);
+end
+if(nargin < 2)
+    depthThresh = 999;
+end
+
+if (length(depthThresh) == 1)
+    depthThresh = 1:depthThresh;
+end
+    
+linkList = aglomCluster(data);
+linkList = depthCheck(linkList);
+listSize = size(data,1);
+
+% linkList(:,4) = 0;
+featureList = cell(listSize-1,3);
+currentRow = [2*listSize-1];
+
+%%
+while (~isempty(currentRow))
+    if(currentRow(1) > listSize)
+        row = currentRow(1) - listSize
+%         rD = linkList(row,4);
+        if any(linkList(row,4)==depthThresh)
+            classList = traceLinkageToBinary(linkList, row);
+            featureList{row,1} = rfFeatureSelection(data(classList>0,:), classList(classList>0));
+            featureList{row,2} = linkList(row,4);
+            featureList{row,3} = fitctree(data(classList>0,featureList{row,1}),classList(classList>0),'PredictorNames',featureNames(featureList{row,1}));
+        end
+        currentRow = [currentRow; linkList(row,1); linkList(row,2)];
+    end
+    currentRow = currentRow(2:end);
+    save('partialResults.mat');
+end
+
+end
\ No newline at end of file