Initializing help system before first use

Decision Tree User Interface

The decision tree user interface consists of a panel on the left to add and delete variables used as predictors and profiles, plus view the properties of the tree, and a toolbar at the top. The VIEWS menu contains canvases to: create and edit the decision tree (known as the Tree View); profile variables (known as the Profile View); analyze treatments using leaf node statistics (known as the Leaf View); and view and analyze a summary of treatments (known as the Treatments Summary view).

In the Tree View, you can manually create or edit splits, and create and assign treatments.

With the menu on the left, you can:

  • expand panels to view the properties of the attachment.
  • view and select Predictors as candidate split variables for the decision tree.
  • view and select Target variables that will be available for profiling.
  • view and add selected Profile variables for profiling the decision tree, to observe how input records are processed at each node — When you profile a decision tree, you choose one or more variables to follow through as the input data is processed.
When you select a target variable for the decision tree, the target automatically becomes a profile variable. Only categorical variables are available for selection as target variables, see Identifying Categorical and Continuous Variables. The target variable can be selected in the Target panel.

Adding and Removing Target, Predictors, and Profile Variables

At the top of the Target, Profiles, and Predictors panels you can add, delete, and search. You can also search for predictors or profile variables listed in their respective panels.

  • The Targets panel lists the variables that will be used in profiling.

  • The Predictors panel lists the predictors you selected as candidate split variables for the decision tree.
  • The Profiles panel lists the profile variables you selected for profiling in the decision tree.
You expand and collapse a panel by clicking the panel column heading.

Editing the Panel Target, Predictor, or Profile variables

Selecting a predictor, profile, or target variable. The panels contain two tool buttons.

Button Description
Open the Configure targets, Choose predictors or Choose variables to profile dialog box
Remove an selected target, predictor, or profile variable from the panel.
Note A target, predictor, or profile that is allocated as a level in the decision tree cannot be removed from the panel until it has been removed from the tree. The button will be greyed out if the selected item is used in the tree.

Targets panel: Click the Targets heading to expand the panel. Click the edit button to display the Configure targets dialog box, where you can add or remove one or more targets for the decision tree. If targets are not used in the decision tree, you can remove them from the panel. Shift-click or Command/Control-click to select multiple targets and then click the remove button. Removing a Target only removes it from the panel and does not remove it from the dataset.

Predictors panel: Click the Predictors heading to expand the panel. Click the edit button in the panel to display the Choose Predictors dialog box, where you can add or remove one or more candidate split variables for the decision tree. If predictors are not used in the decision tree, you can remove them from the panel. Shift-click or Command/Control-click to select multiple predictors and then click the remove button. Removing a predictor only removes it from the panel and does not remove it from the dataset.

Profiles panel: Click the Profiles heading to expand the panel. Click the edit button in the panel to display the Choose variables to profile dialog box, where you can add or remove one or more profile variables to view statistical information for each tree node. You can also remove one or more profile variables from the panel. Shift-click or Command/Control-click to select multiple profile variables and then click the remove button. Removing a profile variable only removes it from the panel and does not remove it from the dataset.

You can also search for predictors or profile variables listed in their respective panels.

(Optional) You can display the Used column. By default, this columns is hidden. The Used column can only appear in the Predictors panel. In the side bar, click Columns to open the Columns Tool panel. Select the Used check box and click Columns again to close the Columns Tool panel.
Note A Y appears in the Used column for all predictors that are levels in the decision tree. These levels can be empty or contain one or more decision nodes.

Navigating the Decision Tree User Interface

The following figure depicts part of a sample authorization application decision tree with:
  • Several treatments
  • Four levels, including the start level
  • Several splits
  • A selected node

Tree View

Tree View

The labeled decision tree elements are:
  1. View controls how your decision tree is displayed.
    • TREE (default): Settings to help you refine your tree.

      The TREE view contains a second options menu, available under the chevron icon.

      This menu contains the following settings:
      • Template
        • Compact—Displays node conditions and treatments without statistics (default view).
        • Counts—Displays node conditions and treatments, as well as the total number of counts and percentage of the total. The Node Stats Options menu also enables the option to swap the displayed values between:
          • weighted (W#T) or raw (#T) count values
          • weighted (W%T) or raw (%T) percent values
          and to optionally display the respective node labels for each setting.
      • Layout

        View the Decision Tree as either a Leveled or Unleveled tree.

      • Color By
        1. No Color (default)—Do not display a color.
        2. Treatments—Each node displays the weighted percentage of the total number of records assigned a treatment. The colors can be edited via the Assign Treatments dialogue, for more see Treatments and Decision Trees. The data in the decision tree must be counted to show this information.

      You can view the distribution by color for each node in the COUNTS and COUNTS-LABEL views by selecting:

    • PROFILE or LEAF—Profile your tree variables and analyze treatments using leaf node statistics.
      Note To use these views, you need to specify which of your tree variables are to be used as profiling variables. For more information on profile variables, see the previous topic Setting up Data Roles for Account Input Data, and Choosing Profiling Statistics later in this topic.
    • TREATMENTS: Profile the treatments and provides aggregated count and weighted count per treatment.

  2. The decision tree toolbar contains commands that let you edit, cut, copy, paste, or delete decision tree elements. Additionally, you can change the magnification of the decision tree by zooming in, out, or to fit the canvas.

    EDIT displays the options to edit the tree. Possible actions are:
    • Trigger recount to update the statistics available for the various views.
    • Compute best split (Available when Target variable is set) Automatically determine and include the best split for a node.
    • Compute percentile split to divide the data for a single numeric predictor node into the number of specified splits (available when there is no target variable.)
    • Insert split to add a condition node with branches. A level is automatically created when you insert a split.
    • Edit split to modify the conditions of the split.
    • Insert level if you know the variables you want to use, in what order, and you want to capture this design before deciding the splits.
    • Cut subtree to cut all nodes below the selected one.
    • Copy subtree to copy all nodes below the selected one.
    • Clear subtree to remove all nodes below the selected one
    • Collapse subtree hides all nodes and edges below the selected node, and displays an arrow icon to indicate more information is available — Click the arrow to reveal the subtree.
    • Collapse entire tree hides all nodes and edges below the Start node, and displays an arrow icon to indicate more information is available — Click the arrow to reveal the subtree.
  3. Toolbar:
    Zoom in
    Zoom out
    Zoom to fit
    Export the tree as FSML, FJDT, PMML, or SAS format file
    Print
    Recount
    Add, Delete and Modify Treatments
    Best Split

    Automatically determine and include the best split for a node, for more see Determining Best Split later in this topic.

    Explore Recommended Splits (Available when Target variable is set)

    Suggest splitting values, for more see Determining Best Split later in this topic.

    Undo (enabled when a change has been made)
    Redo (enabled when an undo has been performed)
    Settings
  4. Level: Each vertical band in the tree, known as a level, corresponds to a variable marked as a Possible tree variable in the Data Roles > Configure Roles for Input Data pane (except the Start level and the Treatments level).
  5. Condition node: A conditional branching point in a decision tree. This node is displayed as a circle in Compact view or as a rectangle containing statistics in the Counts or Counts-label views.
  6. Treatments: The column furthest right in a decision tree displays the color-coded treatments applied to the end or leaf nodes.
Parent Topic Manipulating Decision Trees