def __create_node__(self, parent_node, parent_attribute_val, xs, ys,
attributes, level):
if level >= self.max_depth:
val = self.__get_most_frequent_value__(ys)
return Leaf(parent_attribute_val, val)
distinct_ys = set(ys)
if len(distinct_ys) == 1:
return Leaf(parent_attribute_val, ys[0])
best_attribute_name = self.__find_best_attribute__(
attributes, parent_node.attribute_name, parent_attribute_val, xs,
ys)
new_node = Node(best_attribute_name, parent_attribute_val, parent_node,
self.fn_attribute_val_extractor)
# Pivot by value to segment dataset for children
xs_by_val, ys_by_val = self.__partition_by_attribute__(
best_attribute_name, parent_node.attribute_name,
parent_attribute_val, xs, ys)
# If only one grouping, no further attributes are useful, take majority vote for class
if len(ys_by_val) == 1:
val = self.__get_most_frequent_value__(ys)
return Leaf(parent_attribute_val, val)
# Remove matched attribute
remaining_attributes = [
a for a in attributes if a != best_attribute_name
]
for attr_val in xs_by_val.keys():
xs4val = xs_by_val[attr_val]
ys4val = ys_by_val[attr_val]
child_node = self.__create_node__(new_node, attr_val, xs4val,
ys4val, remaining_attributes,
level + 1)
new_node.add_child(child_node)
return new_node
