Decision Tree Classification - CART

Uses the MADlib built-in function tree_train() to generate a decision tree that predicts the value of a categorical column based on several independent columns.

Information at a Glance

Parameter

 Description
Category Model
MADlib version 1.8+
Data source type DB
Send output to other operators Yes
Data processing tool n/a

The generated tree is a binary tree, with each node representing either a branching condition or a predicted value. The output of the operator can be sent to a predictor or confusion matrix. MADlib 1.8 or higher must be installed on the database.

Important: This operator does not work with MADlib 1.7.1 or lower, due to a change in the way MADlib handles column names. If you have an older version of MADlib, consider using Decision Tree - MADlib instead.

For more information about working with decision trees, see Classification Modeling with Decision Tree.

Input

The input table must have a single, categorical (string or integer) column to predict, and one or more independent columns to serve as input.

Bad or Missing Values
In the source table, any rows that contain null values for the predicted or independent columns are ignored.

Restrictions

  • This operator works only on databases with MADlib 1.8+ installed.
  • Source data tables must have a numeric ID column that uniquely identifies each row in the source table.
  • The prediction column must be numeric, and all predictions are double-precision values.

Configuration

Parameter Description
Notes Notes or helpful information about this operator's parameter settings. When you enter content in the Notes field, a yellow asterisk appears on the operator.
MADlib Schema Name of schema where MADlib is installed. By default, this is madlib.
Model Output Schema The name of the schema to use for MADlib-generated output tables.
Model Output Table The name of the MADlib-generated output table. This table is generated by the tree trainer. An additional table with the same name and the suffix _summary also is generated.
Drop If Exists
  • If Yes (the default), drop the existing table of the same name and create a new one.
  • If No, stop the flow and alert the user that an error has occurred.
ID column All source tables must have a numeric ID column to uniquely identify each row.
Dependent Variable The name of the numeric column to predict. This must be a floating-point column.
Feature List Click Select Columns to specify one or more columns to use as independent variables to predict the dependent variable. See Select Columns dialog for more information.
Split Criterion The algorithm to use for calculating branch nodes during tree generation. For categorical tables, this must be gini, entropy, or misclassification. The default is gini.
Maximum Tree Depth The generated tree does not exceed this depth. If not specified, the default is 10.
Minimum Observations Before Splitting If not specified, the default is 20.
Minimum Observations in Terminal Nodes If not specified, the default is the minimum observations before splitting, divided by 3.
Number of Bins for Split Boundaries If not specified, the default is 100.

Outputs

Visual Output

This operator produces the following tabs.

  • Decision Tree Text - Contains a text representation of the generated decision tree. Each branch node contains a number of rows and a prediction. Branch nodes also contain a branching condition.
  • Decision Tree Graph - Contains a tree graph. Branches reflect split conditions and associated predictions.

Additional Notes

Output to Succeeding Operators
Connect this operator to the following succeeding operators.
  • Prediction
  • Confusion Matrix

Example