TreeAdapter: Return [mean, variance] for values

Author: pavlin-policar

Orange/widgets/visualize/owtreeviewer.py

@@ -376,8 +376,7 @@ def update_node_info_cls(self, node):
     def update_node_info_reg(self, node):
         """Update the printed contents of the node for regression trees"""
         node_inst = node.node_inst
-        # TODO calculate variance in tree adapter
-        mean, var = self.tree_adapter.get_distribution(node_inst)[0][0], 0.
+        mean, var = self.tree_adapter.get_distribution(node_inst)[0]
         insts = len(self.tree_adapter.get_instances_in_nodes(self.dataset, [node_inst]))
         text = "{:.1f} ± {:.1f}<br/>".format(mean, var)
         text += "{} instances".format(insts)
@@ -427,8 +426,8 @@ def toggle_node_color_reg(self):
                 node.backgroundBrush = QBrush(colors[fact * (node_mean - minv)])
         else:
             nodes = list(self.scene.nodes())
-            # TODO Get variance from tree adapter
-            variances = [node.node_inst.value[1] for node in nodes]
+            variances = [self.tree_adapter.get_distribution(node.node_inst)[0][1]
+                         for node in nodes]
             max_var = max(variances)
             for node, var in zip(nodes, variances):
                 node.backgroundBrush = QBrush(def_color.lighter(
@@ -450,7 +449,7 @@ def test():
     a = QApplication(sys.argv)
     ow = OWTreeGraph()
     # data = Table("iris")
-    data = Table("housing")
+    data = Table("housing")[:30]
     clf = SklTreeRegressionLearner()(data)
     # clf = TreeLearner()(data)
     clf.instances = data

Orange/widgets/visualize/owtreeviewer2d.py

@@ -454,7 +454,6 @@ def toggle_line_width(self):
         if self.root_node is None:
             return
 
-        model = self.model
         tree_adapter = self.root_node.tree_adapter
         root_instances = len(tree_adapter.get_instances_in_nodes(
             self.dataset, [self.root_node.node_inst]))

Orange/widgets/visualize/utils/tree/skltreeadapter.py

@@ -66,7 +66,15 @@ def __right_child(self, node):
         return self._tree.children_right[node]
 
     def get_distribution(self, node):
-        return self._tree.value[node]
+        value = self._tree.value[node]
+        # If regression tree, we have to compute variance by hand, we can
+        # detect this because you can't have classification trees when there's
+        # only one class
+        if value.shape[1] == 1:
+            var = np.var(self.get_instances_in_nodes(self.model.instances, node).Y)
+            variances = np.array([(var * np.ones(value.shape[0]))]).T
+            value = np.hstack((value, variances))
+        return value
 
     def get_impurity(self, node):
         return self._tree.impurity[node]