Package biz.k11i.xgboost.gbm

Class GBTree

java.lang.Object

biz.k11i.xgboost.gbm.GBBase

biz.k11i.xgboost.gbm.GBTree

All Implemented Interfaces:

GradBooster, java.io.Serializable

Direct Known Subclasses:

Dart

public class GBTree
extends GBBase

Gradient boosted tree implementation.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
(package private) static class	GBTree.ModelParam

Nested classes/interfaces inherited from interface biz.k11i.xgboost.gbm.GradBooster

GradBooster.Factory

Field Summary

Fields

Modifier and Type	Field	Description
(package private) RegTree[][]	_groupTrees
(package private) GBTree.ModelParam	mparam
private RegTree[]	trees

Fields inherited from class biz.k11i.xgboost.gbm.GBBase

num_class, num_feature, num_output_group

Constructor Summary

Constructors

Constructor	Description
GBTree()

Method Summary

Modifier and Type	Method	Description
RegTree[][]	getGroupedTrees()
void	loadModel(PredictorConfiguration config, ModelReader reader, boolean with_pbuffer)	Loads model from stream.
(package private) float	pred(FVec feat, int bst_group, int root_index, int ntree_limit, float base_score)
private long	predBufferSize()
float[]	predict(FVec feat, int ntree_limit, float base_score)	Generates predictions for given feature vector.
int[]	predictLeaf(FVec feat, int ntree_limit)	Predicts the leaf index of each tree.
java.lang.String[]	predictLeafPath(FVec feat, int ntree_limit)	Predicts the path to leaf of each tree.
float	predictSingle(FVec feat, int ntree_limit, float base_score)	Generates a prediction for given feature vector.

Methods inherited from class biz.k11i.xgboost.gbm.GBBase

setNumClass, setNumFeature

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

mparam

GBTree.ModelParam mparam

trees

private RegTree[] trees

_groupTrees

RegTree[][] _groupTrees

Constructor Detail

GBTree

public GBTree()

Method Detail

loadModel

public void loadModel(PredictorConfiguration config,
                      ModelReader reader,
                      boolean with_pbuffer)
               throws java.io.IOException

Description copied from interface: GradBooster

Loads model from stream.

Parameters:

config - predictor configuration

reader - input stream

with_pbuffer - whether the incoming data contains pbuffer

Throws:

java.io.IOException - If an I/O error occurs

predict

public float[] predict(FVec feat,
                       int ntree_limit,
                       float base_score)

Description copied from interface: GradBooster

Generates predictions for given feature vector.

Parameters:

feat - feature vector

ntree_limit - limit the number of trees used in prediction

base_score - base score to initialize prediction

Returns:

prediction result

predictSingle

public float predictSingle(FVec feat,
                           int ntree_limit,
                           float base_score)

Description copied from interface: GradBooster

Generates a prediction for given feature vector.

This method only works when the model outputs single value.

Parameters:

feat - feature vector

ntree_limit - limit the number of trees used in prediction

base_score - base score to initialize prediction

Returns:

prediction result

pred

float pred(FVec feat,
           int bst_group,
           int root_index,
           int ntree_limit,
           float base_score)

predictLeaf

public int[] predictLeaf(FVec feat,
                         int ntree_limit)

Description copied from interface: GradBooster

Predicts the leaf index of each tree. This is only valid in gbtree predictor.

Parameters:

feat - feature vector

ntree_limit - limit the number of trees used in prediction

Returns:

predicted leaf indexes

predictLeafPath

public java.lang.String[] predictLeafPath(FVec feat,
                                          int ntree_limit)

Description copied from interface: GradBooster

Predicts the path to leaf of each tree. This is only valid in gbtree predictor.

Parameters:

feat - feature vector

ntree_limit - limit the number of trees used in prediction

Returns:

predicted path to leaves

predBufferSize

private long predBufferSize()

getGroupedTrees

public RegTree[][] getGroupedTrees()

Returns:

A two-dim array, with trees grouped into classes.