Decision Trees

This library provides support for creating and applying decision trees.

Basic Usage

For an example of how to use this library, see the Gradient Boosted Decision Trees tutorial.

Interface

This library includes support both for decision trees with integer features and labels (DecisionTree) and for decision trees with real-valued features and labels (DecisionTreeReal).

Types

• type label
  • Labels (the decisions that a decision tree makes) are either int or real.
• type feature
  • Individual features are either int * int or int * real, where the left hand side is the feature ID and the right hand side is the feature value.
• type features
  • Feature vectors are either an int array or real array, where the index into the array is the feature ID.
• type t
  • DecisionTree.t and DecisionTreeReal.t are the types of the decision trees themselves.

Methods

• val lessFeature: features * (int * label) -> bool
• val forward: t * features -> label
• val makeLf: label -> t
• val makeNd: t * feature * t -> t
• val toString: t -> string
• val split: (features * label) list * (int * label) -> ((features * label) list) * ((features * label) list)
• val recordsToString: (features * label) list -> string
• val leafNum: t -> int

Method Overview

• lessFeature (ft, (featureId, featureValue))
  • Determine whether the value of the given feature is less than the feature value in the feature vector ft.
• forward (dt, ft)
  • Run the decision tree dt to produce a label for the feature vector ft.
• makeLf label
  • Creates a decision tree that is just a leaf node that always returns the same label label.
• makeNd (dt1, feature, dt2)
  • Creates a decision tree that uses the left tree to get a label if feature vector being processed is less than feature and uses the right tree otherwise.
• toString dt
  • Converts a decision tree to a string.
• split (data, (featureId, featureValue))
  • Splits a list of input data by a feature. The input data is a collection of feature vectors and labels.
• recordsToString data
  • Converts a list of feature vectors and their labels to a string.
• leafNum dt
  • Returns the number of leaves in a decision tree.