A data structure that models a set of specific questions, their possible answers, and the actions that follow from a series of answers.
- iOS 10.0+
- macOS 10.12+
- Mac Catalyst 13.0+
- tvOS 10.0+
You can define a decision tree manually, by specifying questions, answers, and actions, or you can allow the
GKDecision class to automatically learn a predictive model based on example data. A decision tree has several elements:
Attributes represent individual questions to be answered or choices to be made.
Branches are the possible answers to the questions or choices posed by each attribute.
Actions are the final outcomes of the tree’s decision-making process. Each branch from an attribute leads either to another attribute or to an action.
When you use the
GKDecision class, attributes and actions can be any object type relevant to your app or game. You can define branches for specific answer values, using predicates, or with weights that influence a random decision. For example, a strategy combat game might use a decision tree to choose what a character should do on its turn, based on several criteria about the match in progress. In this case:
For attributes, you might use (non-user-visible) strings that represent those criteria, such as
"Type?"(what type of enemy is the character’s opponent?),
"HP?"(how much health does the opponent have remaining?), and
"Special?"(is the character’s special move available for use?).
For branches, you’d use an appropriate style for each attribute. The
"Type?"attribute might have a branch for each possible enemy type, but the
"HP?"attribute could use predicates to determine whether the enemy’s health is above or below a certain threshold value.
For actions, you might define your own enumerated type representing the kinds of attacks the character can choose (such as
Barrier). Alternately, you might use instances of your own custom classes representing items or spells available to the character.
Figure 1 illustrates a possible tree structure based on the above example attributes, branches, and actions.
Creating a Decision Tree
GKDecision class offers two ways to create decision trees.
In a manually defined decision tree, you define each attribute to be tested (or question to be asked), the possible branches (or answers) from each attribute, and the actions (or final outcomes) resulting from each complete series of attribute tests and branches. To manually create a tree, start by using the
init initializer to define the first question to be asked. Then, use
GKDecision methods on the new tree’s
root object to define branches and the child nodes they lead to, with accompanying attribute or action values.
In a learned decision tree, you provide a set of attributes (or questions); a body of example items, each of which represents a set of attribute values (or answers to questions); and the final action to be taken for each example. The
GKDecision class then automatically infers a decision tree structure that, when presented with a set of attribute values matching or similar to one of your examples, predicts the corresponding action. To create a learned decision tree, use the
init initializer. Table 1 shows sample input for a learned decision tree (based on the same hypothetical game shown in Figure 1).
Can Use Special Move
Move to Use
After creating either kind of decision tree, you can use the inherited
description property to examine its structure.
After you’ve created a tree, use the
find method to evaluate it and choose an action. When you call that method, you provide a set of inputs (values for attributes, or answers to questions), and the tree follows the branches corresponding to each input value to produce an action.