Merge branch 'aiim-research:main' into main

xsk07 · Aug 28, 2024 · 27f4d91 · 27f4d91
2 parents dd2860a + 8c05cbb
commit 27f4d91
Show file tree

Hide file tree

Showing 10 changed files with 753 additions and 124 deletions.
diff --git a/lab/1-evaluation_pipeline.ipynb b/lab/1-evaluation_pipeline.ipynb
diff --git a/src/data_analysis/future/data_analyzer.py b/src/data_analysis/future/data_analyzer.py
@@ -265,10 +265,10 @@ def draw_graph(cls, data_instance, position=None, img_store_address=None):
             layout = nx.spring_layout
             position = layout(G)
 
-        edge_colors = ['cyan' for u, v in G.edges()]
-        node_colors = ['cyan' for node in G.nodes()]
+        edge_colors = ['gray' for u, v in G.edges()]
+        node_colors = ['gray' for node in G.nodes()]
 
-        nx.draw_networkx(G=G, pos=position, node_color=node_colors, edge_color=edge_colors, with_labels=True)
+        nx.draw_networkx(G=G, pos=position, node_color=node_colors, edge_color=edge_colors, with_labels=False, node_size=100)
 
         if img_store_address:
             plt.savefig(img_store_address, format='svg')
@@ -305,9 +305,9 @@ def draw_counterfactual_actions(cls,
 
         # Add shared nodes and edges in grey
         for node in nodes_shared:
-            G.add_node(node, color='cyan')
+            G.add_node(node, color='gray')
         for edge in edges_shared:
-            G.add_edge(*edge, color='cyan')
+            G.add_edge(*edge, color='gray')
 
         # Add deleted nodes and edges in red
         for node in nodes_deleted:
@@ -324,7 +324,7 @@ def draw_counterfactual_actions(cls,
         edge_colors = [G[u][v]['color'] for u, v in G.edges()]
         node_colors = [G.nodes[node]['color'] for node in G.nodes()]
 
-        nx.draw_networkx(G=G, pos=position, node_color=node_colors, edge_color=edge_colors, with_labels=True)
+        nx.draw_networkx(G=G, pos=position, node_color=node_colors, edge_color=edge_colors, with_labels=False, node_size=100)
 
         if img_store_address:
             plt.savefig(img_store_address, format='svg')

diff --git a/src/dataset/instances/graph.py b/src/dataset/instances/graph.py
@@ -86,6 +86,10 @@ def num_edges(self):
     @property
     def num_nodes(self):
         return len(self.data)
+
+    @property
+    def is_directed(self):
+        return self.directed
 
     def nodes(self):
         return [ i for i in range(self.data.shape[0])]

diff --git a/src/evaluation/future/evaluator.py b/src/evaluation/future/evaluator.py
@@ -180,4 +180,10 @@ def pickle_explanations(self, store_path):
 
         # Pickle the list into the specified file
         with open(pickle_file_path, 'wb') as pickle_file:
-            pickle.dump(self._explanations, pickle_file)
+            for exp in self._explanations:
+                exp.input_instance._dataset = None
+                for inst in exp.counterfactual_instances:
+                    inst._dataset = None
+
+            inst_cf_pairs = [(exp.input_instance, exp.counterfactual_instances[0]) for exp in self._explanations]
+            pickle.dump(self.inst_cf_pairs, pickle_file)
diff --git a/src/explainer/future/ensemble/aggregators/base.py b/src/explainer/future/ensemble/aggregators/base.py
@@ -123,7 +123,7 @@ def get_all_edge_differences(self, instance: DataInstance, explanations: List[Da
         edges = [(row, col) for row, col in zip(*np.where(edge_change_freq_matrix))]
 
         # If we are working with directed graphs
-        if instance.directed:
+        if instance.is_directed:
             filtered_edges = edges
         else: # if we are working with undirected graphs
             filtered_edges = []

diff --git a/src/explainer/future/ensemble/aggregators/bidirectional.py b/src/explainer/future/ensemble/aggregators/bidirectional.py
@@ -114,7 +114,7 @@ def oblivious_forward_search(self, instance, e_add, e_rem, k=5, maximum_oracle_c
                         # remove
                         i,j = e_rem.pop(0)
 
-                        if instance.directed:
+                        if instance.is_directed:
                             cf_candidate_matrix[i][j]=0
                         else:
                             cf_candidate_matrix[i][j]=0
@@ -128,7 +128,7 @@ def oblivious_forward_search(self, instance, e_add, e_rem, k=5, maximum_oracle_c
                         # add
                         i,j = e_add.pop(0)
 
-                        if instance.directed:
+                        if instance.is_directed:
                             cf_candidate_matrix[i][j]=1
                         else:
                             cf_candidate_matrix[i][j]=1
@@ -173,7 +173,10 @@ def oblivious_backward_search(self, instance, cf_instance, changed_edges, k=5, m
             # Revert the changes on the selected edges
             for i,j in edges_i:
                 gci[i][j] = abs(1 - gci[i][j])
-                gci[j][i] = abs(1 - gci[j][i])
+
+                # If the graph is undirected we need to undo the symmetrical edge too 
+                if not instance.is_directed:
+                    gci[j][i] = abs(1 - gci[j][i])
 
             reduced_cf_inst = GraphInstance(id=instance.id, label=0, data=gci)
             self.dataset.manipulate(reduced_cf_inst)

diff --git a/src/explainer/future/ensemble/aggregators/frequency.py b/src/explainer/future/ensemble/aggregators/frequency.py
@@ -52,6 +52,11 @@ def real_aggregate(self, explanations: List[LocalGraphCounterfactualExplanation]
             if mod_freq_matrix[edge[0], edge[1]] >= freq_threshold:
                 intersection_matrix[edge[0], edge[1]] = abs(intersection_matrix[edge[0], edge[1]] - 1 )
 
+                # If the graphs are undirected
+                if not input_inst.is_directed:
+                    # Assign to the symetrical edge the same value than to the original edge
+                    intersection_matrix[edge[1], edge[0]] = intersection_matrix[edge[0], edge[1]] # The original edge was already modified
+
         # Create the aggregated explanation
         aggregated_instance = GraphInstance(id=input_inst.id, label=1-input_inst.label, data=intersection_matrix)
         self.dataset.manipulate(aggregated_instance)

diff --git a/src/explainer/future/ensemble/aggregators/intersection.py b/src/explainer/future/ensemble/aggregators/intersection.py
@@ -27,6 +27,11 @@ def real_aggregate(self, explanations: List[LocalGraphCounterfactualExplanation]
             if mod_freq_matrix[edge[0], edge[1]] == exp_count:
                 intersection_matrix[edge[0], edge[1]] = abs(intersection_matrix[edge[0], edge[1]] - 1 )
 
+                # If the graphs are undirected
+                if not input_inst.is_directed:
+                    # Assign to the symetrical edge the same value than to the original edge
+                    intersection_matrix[edge[1], edge[0]] = intersection_matrix[edge[0], edge[1]] # The original edge was already modified
+
         # Create the aggregated explanation
         aggregated_instance = GraphInstance(id=input_inst.id, label=1-input_inst.label, data=intersection_matrix)
         self.dataset.manipulate(aggregated_instance)

diff --git a/src/explainer/future/ensemble/aggregators/rand.py b/src/explainer/future/ensemble/aggregators/rand.py
@@ -43,12 +43,15 @@ def real_aggregate(self, explanations: List[LocalGraphCounterfactualExplanation]
             adj_matrix = copy.deepcopy(input_inst.data)
             # Randomly sample a number of edges equivalent to the smallest base explanation
             sampled_edges = random.sample(change_edges, min_changes)
-            #TODO consider the case of undirected graphs
 
             # Try to modified the chosen edges one by one until a counterfactual is found
             for edge in sampled_edges:
                 adj_matrix[edge[0], edge[1]] = abs( adj_matrix[edge[0], edge[1]] - 1 )
 
+                if not input_inst.is_directed:
+                    # Assign to the symetrical edge the same value than to the original edge
+                    adj_matrix[edge[1], edge[0]] = adj_matrix[edge[0], edge[1]] # The original edge was already modified
+
                 # Creating an instance with the modified adjacency matrix
                 aggregated_instance = GraphInstance(id=input_inst.id,
                                                        label=0,

diff --git a/src/explainer/future/ensemble/aggregators/union.py b/src/explainer/future/ensemble/aggregators/union.py
@@ -25,6 +25,11 @@ def real_aggregate(self, explanations: List[LocalGraphCounterfactualExplanation]
         for edge in mod_edges:
             union_matrix[edge[0], edge[1]] = abs(union_matrix[edge[0], edge[1]] - 1 )
 
+            # If the graphs are undirected
+            if not input_inst.is_directed:
+                    # Assign to the symetrical edge the same value than to the original edge
+                    union_matrix[edge[1], edge[0]] = union_matrix[edge[0], edge[1]] # The original edge was already modified
+
         # Create the aggregated explanation
         aggregated_instance = GraphInstance(id=input_inst.id, label=1-input_inst.label, data=union_matrix)
         self.dataset.manipulate(aggregated_instance)