EECS_4412_A2 DECISION TREES:

• A simple Decision tree program
• Part 1: creating a function to determine if a node is an Internal or a Leaf node
• Part 2: Creating a function that calculates the entropy (Measure of Impurity in the data) of a node to be used for splitting a node based on the Information Gain criterion
• Part 3: Use the entropy function to calculate the Information Gain of a node to determine which attribute is the best attribute to split on
• Part 4: Use the functions created in Part 2 & 3 to determine the best attribute that has the highest Information Gain so that the values (examples) in the node can be split on that attribute
• Part 5: Splitting the nodes at each level by using the above functions using the information gaain criteria.

  • What the function does:

  • Structure of the variables:

    • The dataframe of the previous level that we will create the children from:

       dataframe_dict = {"Node_1_1": the dataframe at the root node}

    • after first pass:

         dataframe_dict = {
          "Node_2_1": dataframe_2_1,
          "Node_2_2": dataframe_2_2,
          "Node_2_3": dataframe_2_3
      }

    • Attributes remaining after splitting the nodes at each level:

      • remaining_attrs = {"Node_1_1": ["BUYING", "MAINTENANCE", "DOORS", "PERSONS", "LUG_BOOT", "SAFETY"]}

      • after first pass (assuming we're splitting on attr "PERSONS"):

      •   remaining attrs = {
            "Node_2_1": ["BUYING", "MAINTENANCE", "DOORS", "LUG_BOOT", "SAFETY"],
           "Node_2_2": ["BUYING", "MAINTENANCE", "DOORS", "LUG_BOOT", "SAFETY"],
           "Node_2_3": ["BUYING", "MAINTENANCE", "DOORS", "LUG_BOOT", "SAFETY"]
        }

    • The tree model which contains information about the current state of the tree:

      • tree_model = [tree_connectivity, node_types, node_labels]

    • visual rep. of what will be store in the tree_model:
    • After first pass: