DecisionTree Class Reference

#include <DecisionTree.h>

Inheritance diagram for DecisionTree:

Public Member Functions
	DecisionTree (int min_samples_per_split=2, int max_depth=10, ImpurityMeasure decision_method=ImpurityMeasure::GINI)
double	impurity (const Matrix< double > &x)
double	information_gain (const Matrix< double > &parent, const Matrix< double > &left_child, const Matrix< double > &right_child)
struct DataSplit	best_split (const Matrix< double > &X, const Matrix< double > &y)
DecisionNode *	GetRootNode () const
void	fit (const Matrix< double > &X, const Matrix< double > &y) override
Matrix< double >	predict (const Matrix< double > &X) override
Matrix< double >	transform (const Matrix< double > &in) override
virtual void	fit (const Matrix< double > &X, const Matrix< double > &y)=0
virtual Matrix< double >	predict (const Matrix< double > &)=0
virtual Matrix< double >	transform (const Matrix< double > &)=0

Private Member Functions
DecisionNode *	build_tree (const Matrix< double > &X, const Matrix< double > &y, int depth=0)
double	_predict (const Matrix< double > &x, DecisionNode *node)
void	render_node (std::ostream &ostr, DecisionNode *node, const char *side, int depth) const

Private Attributes
ImpurityMeasure	_decision_method
int	_max_depth = 10
int	_min_samples_per_split = 2
DecisionNode *	base_node

Friends
std::ostream &	operator<< (std::ostream &ostr, const DecisionTree &tree)

Detailed Description

Decision (Bi-)Tree implementation

• The constructor allows providing a decision_method for the impurity measure that is used when selecting a good split
• The information_gain(parent, left_child, right_child) calculates the information gain value of a split between a parent and two children
• The _best_split(X, y) function calculates the best splitting parameters for input features X and a target variable y
  • It does so by iterating over every column in X and every thresold value in every column to find the optimal split using information gain
• The _build(X, y, depth) function recursively builds a decision tree until the data is _non splittable
• The fit(X, y) function calls the _build() function and stores the built tree to the constructor
• The _predict(x) function traverses the tree to classify a single instance
• The predict(X) function applies the _predict() function to every instance in matrix X

Constructor & Destructor Documentation

DecisionTree()

DecisionTree::DecisionTree ( int  min_samples_per_split = 2, int  max_depth = 10, ImpurityMeasure  decision_method = ImpurityMeasure::GINI  )

inline

Member Function Documentation

_predict()

double DecisionTree::_predict ( const Matrix< double > &  x, DecisionNode *  node  )

inlineprivate

Method to recursively traverse the tree to perform a prediction

Parameters

x	input value(s)
node	the current node to evaluate

Returns

: prediction for given node

best_split()

struct DataSplit DecisionTree::best_split ( const Matrix< double > &  X, const Matrix< double > &  y  )

inline

build_tree()

DecisionNode * DecisionTree::build_tree ( const Matrix< double > &  X, const Matrix< double > &  y, int  depth = 0  )

inlineprivate

Method to recursively build decision tree nodes.

Uses pruning to decide to split the data further. Class-Hyperparameters:

• max_depth
• min_samples_per_split

Parameters

X	input data of current node
y	related class labels

Returns

: A DecisionNode instance holding the next node.

fit()

void DecisionTree::fit ( const Matrix< double > &  X, const Matrix< double > &  y  )

inlineoverridevirtual

Implements training algorithm

Parameters

X	array-like with the shape: [n_samples, n_features]
y	array-like with shape: [n_samples, 1]

Returns

this

Implements Predictor.

GetRootNode()

DecisionNode * DecisionTree::GetRootNode ( ) const

inline

impurity()

double DecisionTree::impurity ( const Matrix< double > &  x )

inline

information_gain()

double DecisionTree::information_gain ( const Matrix< double > &  parent, const Matrix< double > &  left_child, const Matrix< double > &  right_child  )

inline

Calculates the information gain of a set of values given a split of two children.

Note: Pass your labels in here, not your input data!

Parameters

parent	super set of label values
left_child	left split of parent set
right_child	right split of parent set

Returns

information gain of given split

predict()

Matrix< double > DecisionTree::predict ( const Matrix< double > &  X )

inlineoverridevirtual

Makes prediction for given input

Returns

Implements Predictor.

render_node()

void DecisionTree::render_node ( std::ostream &  ostr, DecisionNode *  node, const char *  side, int  depth  ) const

inlineprivate

transform()

Matrix< double > DecisionTree::transform ( const Matrix< double > &  in )

inlineoverridevirtual

Implements Predictor.

Friends And Related Function Documentation

operator<<

std::ostream & operator<< ( std::ostream &  ostr, const DecisionTree &  tree  )

friend

Member Data Documentation

_decision_method

ImpurityMeasure DecisionTree::_decision_method

private

_max_depth

int DecisionTree::_max_depth = 10

private

_min_samples_per_split

int DecisionTree::_min_samples_per_split = 2

private

base_node

DecisionNode* DecisionTree::base_node

private

---

The documentation for this class was generated from the following file:

• include/math/ds/DecisionTree.h