Merge pull request #1515 from softmatterlab/develop

BRAPH 2.0.0.a3

CITATION.cff

@@ -26,7 +26,7 @@ authors:
     orcid: 'https://orcid.org/0000-0002-5388-7712'
   - given-names: Anna
     family-names: Canal-Garcia
-    email: [email protected]
+    email: anna.canal.garcia@ki.se
     affiliation: Karolinska Institute
     orcid: 'https://orcid.org/0000-0001-7460-741X'
   - given-names: Yu-Wei

braph2genesis/measures/_EdgeOverlap.gen.m

@@ -21,6 +21,7 @@
 MultiplexBUT
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_MultiRichClubDegree.gen.m

@@ -23,6 +23,7 @@
 MultiplexBUT
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props!
 

braph2genesis/measures/_MultiRichClubStrength.gen.m

@@ -19,6 +19,7 @@
 %%% ¡compatible_graphs!
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props!
 

braph2genesis/measures/_MultiplexClustering.gen.m

@@ -23,6 +23,7 @@
 MultiplexBU
 MultiplexBUD
 MultiplexBUT
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_MultiplexClusteringAv.gen.m

@@ -22,6 +22,7 @@
 MultiplexBU
 MultiplexBUD
 MultiplexBUT
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_MultiplexCorePeriphery.gen.m

@@ -24,6 +24,7 @@ the default coefficients are (1/layernumber).
 MultiplexBUT
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_MultiplexKCore.gen.m

@@ -19,6 +19,7 @@
 MultiplexBU
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props!
 %%% ¡prop! 

braph2genesis/measures/_MultiplexKCorenessCentrality.gen.m

@@ -19,6 +19,7 @@ but not to the (k+1)-core.
 MultiplexWD
 MultiplexBU
 MultiplexBD
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_MultiplexParticipation.gen.m

@@ -24,6 +24,7 @@
 MultiplexBUD
 MultiplexBUT
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_MultiplexParticipationAv.gen.m

@@ -19,6 +19,7 @@
 MultiplexBUD
 MultiplexBUT
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_MultiplexTriangles.gen.m

@@ -24,6 +24,7 @@
 MultiplexBU
 MultiplexBUD
 MultiplexBUT
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_Multirichness.gen.m

@@ -23,6 +23,7 @@ the default coefficients are (1/layernumber).
 MultiplexBUT
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props!
 %%% ¡prop! 

braph2genesis/measures/_OverlappingDegree.gen.m

@@ -19,6 +19,7 @@
 MultiplexBUD
 MultiplexBUT
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_OverlappingDegreeAv.gen.m

@@ -19,6 +19,7 @@
 MultiplexBUD
 MultiplexBUT
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_OverlappingStrength.gen.m

@@ -16,6 +16,7 @@
 
 %%% ¡compatible_graphs!
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_OverlappingStrengthAv.gen.m

@@ -16,6 +16,7 @@
 
 %%% ¡compatible_graphs!
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_PathLength.gen.m

@@ -72,6 +72,7 @@
 end
 value = path_length;
 
+
 %% ¡tests!
 
 %%% ¡test!

braph2genesis/measures/_Richness.gen.m

@@ -27,6 +27,7 @@
 MultiplexBUT
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_Strength.gen.m

@@ -17,6 +17,7 @@
 %%% ¡compatible_graphs!
 GraphWU
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_StrengthAv.gen.m

@@ -17,6 +17,7 @@
 %%% ¡compatible_graphs!
 GraphWU
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_WeightedEdgeOverlap.gen.m

@@ -17,6 +17,7 @@
 %%% ¡compatible_graphs!
 MultiplexWD
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_WeightedMultiplexParticipation.gen.m

@@ -16,6 +16,7 @@
 
 %%% ¡compatible_graphs!
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/measures/_WeightedMultiplexParticipationAv.gen.m

@@ -16,6 +16,7 @@
 
 %%% ¡compatible_graphs!
 MultiplexWU
+OrderedMultiplexWU
 
 %% ¡props_update!
 

braph2genesis/neuralnetworks/_NNClassifierCrossValidation.gen.m

@@ -16,7 +16,7 @@
 %%%% ¡default!
 [100 100]
 %%%% ¡gui!
-pr = PanelPropRVectorSmart('EL', nncv, 'PROP', NNClassifierCrossValidation.LAYERS, 'MAX', 100000, 'MIN', 1, varargin{:});
+pr = PanelPropRVectorSmart('EL', nncv, 'PROP', NNClassifierCrossValidation.LAYERS, 'MAX', 100000, 'MIN', 1, 'UNIQUE_VALUE', false, varargin{:});
 
 %%% ¡prop!
 BATCH (parameter, scalar) is the size of the mini-batch to use for each training iteration.
@@ -226,32 +226,67 @@
 FEATURE_IMPORTANCE (result, cell) is the feature importance obtained with permutation analysis.
 %%%% ¡calculate!
 if ~isa(nncv.get('GR1').getr('SUB_DICT'), 'NoValue')
-    nne_dict = nncv.memorize('NNE_DICT');
-    feature_importances = nne_dict.getItem(1).get('FEATURE_PERMUTATION_IMPORTANCE');
-    if length(feature_importances) == 0
+    if any(ismember(nncv.get('GR1').get('SUB_DICT').getItem(1).get('INPUT_LABEL'), subclasses('Measure', [], [], true)))
         feature_importances = {};
+        if ~braph2_testing
+            questdlg('Feature importance analysis does not apply to the input of graph measures.', ...
+                'User Request', ...
+                'Ok', 'Ok');
+        end
     else
-        for i = 2:1:nne_dict.length()
-            feature_importance = nne_dict.getItem(i).get('FEATURE_PERMUTATION_IMPORTANCE');
-            feature_importances = cellfun(@(x, y) x + y, feature_importances, feature_importance, 'UniformOutput', false);
+        nne_dict = nncv.memorize('NNE_DICT');
+        wb = braph2waitbar(nncv.get('WAITBAR'), 0, 'Analysing feature importance...');
+        feature_importances = nne_dict.getItem(1).get('FEATURE_PERMUTATION_IMPORTANCE');
+        braph2waitbar(wb, .30 + .70 * 1 / nne_dict.length, ['Analysing feature importance for the fold ' num2str(1) ' out of ', num2str(nne_dict.length), '...']);
+        if length(feature_importances) == 0
+            feature_importances = {};
+        else
+            for i = 2:1:nne_dict.length()
+                feature_importance = nne_dict.getItem(i).get('FEATURE_PERMUTATION_IMPORTANCE');
+                feature_importances = cellfun(@(x, y) x + y, feature_importances, feature_importance, 'UniformOutput', false);
+                braph2waitbar(wb, .30 + .70 * i / nne_dict.length, ['Analysing feature importance for the fold ' num2str(i) ' out of ', num2str(nne_dict.length), '...']);
+            end
+            feature_importances = cellfun(@(x) x/nne_dict.length, feature_importances, 'UniformOutput', false);
         end
-        feature_importances = cellfun(@(x) x/nne_dict.length, feature_importances, 'UniformOutput', false);
+        braph2waitbar(wb, 'close')
     end
 else
     feature_importances = {};
 end
 
 value = feature_importances;
+%%%% ¡gui!
+if ~braph2_testing && ~isa(nncv.get('GR').get('SUB_DICT'), 'NoValue')
+    if string(nncv.get('GR1').get('SUB_DICT').getItem(1).get('INPUT_TYPE')) == 'structural_data'
+        pr = PPNNEvaluatorFeatureImportanceStructuralData('EL', nncv, 'PROP', NNClassifierCrossValidation.FEATURE_IMPORTANCE, varargin{:});
+    elseif string(nncv.get('GR1').get('SUB_DICT').getItem(1).get('INPUT_TYPE')) == 'adjacency_matrices'
+        pr = PPNNEvaluatorFeatureImportanceAdjacency('EL', nncv, 'PROP', NNClassifierCrossValidation.FEATURE_IMPORTANCE, varargin{:});
+    else
+        pr = PanelPropCell('EL', nncv, 'PROP', NNClassifierCrossValidation.FEATURE_IMPORTANCE, varargin{:});
+    end
+end
 
 %% ¡props_update!
 
+%%% ¡prop!
+GR (data, item) is a group of NN subjects.
+%%%% ¡settings!
+'NNGroup'
+%%%% ¡postprocessing!
+if ~braph2_testing && isempty(nncv.get('GR').get('SUB_DICT').get('IT_LIST'))
+    nncv.set('GR', nncv.get('GR1'));
+end
+
 %%% ¡prop!
 PFFI (gui, item) contains the panel figure of the feature importance.
 %%%% ¡settings!
 'PFFeatureImportance'
 %%%% ¡postprocessing!
 if ~braph2_testing % to avoid problems with isqual when the element is recursive
-    nncv.memorize('PFFI').set('NNE', nncv, 'PROP', NNClassifierCrossValidation.FEATURE_IMPORTANCE, 'BA', nncv.get('GR1').get('SUB_DICT').getItem(1).get('BA'))
+    nncv.memorize('PFFI').set('NNE', nncv, ...
+        'PROP', NNClassifierCrossValidation.FEATURE_IMPORTANCE, ...
+        'INPUT_TYPE', nncv.get('GR1').get('SUB_DICT').getItem(1).get('INPUT_TYPE'), ...
+        'BA', nncv.get('GR1').get('SUB_DICT').getItem(1).get('BA'))
 end
 %%%% ¡gui!
 pr = PanelPropItem('EL', nncv, 'PROP', NNClassifierCrossValidation.PFFI, ...
@@ -264,7 +299,7 @@
 'NNGroup'
 %%%% ¡calculate!
 if ~isa(nncv.get('GR1').getr('SUB_DICT'), 'NoValue')
-    gr = nncv.get('NNE_DICT').getItem(1).get('GR_PREDICTION');
+    gr = nncv.memorize('NNE_DICT').getItem(1).memorize('GR_PREDICTION');
     gr_prediction = NNGroup( ...
         'ID', 'NN Group with NN prediction', ...
         'LABEL', 'The predictions are obatined from K-fold cross validation', ...

braph2genesis/neuralnetworks/_NNClassifierDNN.gen.m

@@ -20,7 +20,7 @@
     end
 end
 %%%% ¡gui!
-pr = PanelPropRVectorSmart('EL', nn, 'PROP', NNClassifierDNN.LAYERS, 'MAX', 100000, 'MIN', 1, varargin{:});
+pr = PanelPropRVectorSmart('EL', nn, 'PROP', NNClassifierDNN.LAYERS, 'MAX', 100000, 'MIN', 1, 'UNIQUE_VALUE', false, varargin{:});
 
 %%% ¡prop!
 BATCH (parameter, scalar) is the size of the mini-batch to use for each training iteration.
@@ -48,7 +48,7 @@
 1
 
 %%% ¡prop!
-VERBOSE (parameter, logical) is an indicator to display trining progress information.
+VERBOSE (parameter, logical) is an indicator to display training progress information.
 %%%% ¡default!
 false
 
@@ -127,38 +127,48 @@
 end
 
 %% ¡methods!
-function [inputs, num_features] = reconstruct_inputs(nn, gr)
+function [inputs, num_features, masks] = reconstruct_inputs(nn, gr)
 %RECONSTRUCT_INPUTS reconstructs the inputs for NN
 %
 % [INPUTS, NUM_FEATURES] = RECONSTRUCT_INPUTS(NN, GR) reconstructs the
 %   inputs from NN group. According to the tyep of this fully-connected NN,
 %   this function will flatten the input into a vector for each datapoint
 %   and return the number of features for input layer as well.
-
+    
     if gr.get('SUB_DICT').length() == 0
         inputs = [];
         num_features = 0;
     else
-         % get the mask for selecting the relevant input features
+        % get the mask for selecting the relevant input features
         mask_tmp = gr.get('FEATURE_SELECTION_MASK');
-        masks = {};
-        for i = 1:1:length(mask_tmp)
+        masks = cell(size(mask_tmp));
+        for i = 1:1:numel(mask_tmp)
             mask = mask_tmp{i};
+            mask = rescale(mask);
             [~, idx_all] = sort(mask(:), 'descend');
             percentile = nn.get('FEATURE_SELECTION_RATIO');
             num_top_idx = ceil(percentile * numel(mask));
             mask(idx_all(1:num_top_idx)) = 1;
             mask(idx_all(end - (length(idx_all) - num_top_idx - 1):end)) = 0;
             masks{i} = mask;
         end
-
+    
         % apply the mask to select input features
         inputs = [];
         inputs_tmp = gr.get('INPUTS');
         for i = 1:1:length(inputs_tmp)
             input = inputs_tmp{i};
-            input_per_sub = cellfun(@(x, y) x(y == 1), input, masks, 'UniformOutput', false);
-            input_per_sub = cell2mat(input_per_sub');
+            input_per_sub = cellfun(@(x, y) transpose(x(y == 1)), input, masks, 'UniformOutput', false);
+            for j = 1:1:numel(input_per_sub)
+                data = input_per_sub{j};
+                data(isnan(data)) = 0;
+                data(isinf(data)) = 0;
+                if size(data, 1) < size(data, 2)
+                    data = data';
+                end
+                input_per_sub_vector{j} = data;
+            end
+            input_per_sub = cell2mat(input_per_sub_vector');
             inputs = [inputs input_per_sub];
         end
         num_features = length(inputs(:, 1));

braph2genesis/neuralnetworks/_NNClassifierDataSplit.gen.m

@@ -154,7 +154,7 @@
     y = cellfun(@(x) cell2mat(x.get('TARGET')), gr_train.get('SUB_DICT').getItems(), 'UniformOutput', false);
     y = cell2mat(y);
     label = y(1, :);
-    num_feature_cluster = length(data{1});
+    num_feature_cluster = numel(data{1});
     value = cell(size(data{1}));
 
     % setup counter for waitbar
@@ -171,7 +171,7 @@
         if ~isempty(mask)
             for i = 1:numel(mask)
                 data_per_feature = cellfun(@(v)v(i), data_per_cluster);
-                if(any(isinf(data_per_feature)))
+                if(any(isinf(data_per_feature)) || any(isnan(data_per_feature)))
                     mask(i) = 0;
                 else
                     mask(i) = nnds.mutual_information_analysis(data_per_feature, label, 5);