hex.genmodel.algos.tree

Class SharedTreeMojoModel

java.lang.Object

hex.genmodel.GenModel

hex.genmodel.MojoModel

hex.genmodel.algos.tree.SharedTreeMojoModel

All Implemented Interfaces:

IGenModel, java.io.Serializable, IGeneratedModel

Direct Known Subclasses:

DrfMojoModel, GbmMojoModel

public abstract class SharedTreeMojoModel
extends MojoModel

Common ancestor for DrfMojoModel and GbmMojoModel. See also: `hex.tree.SharedTreeModel` and `hex.tree.TreeVisitor` classes.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static interface	SharedTreeMojoModel.DecisionPathTracker<T>
static class	SharedTreeMojoModel.LeafDecisionPathTracker
static class	SharedTreeMojoModel.LeafNodeAssignments
static class	SharedTreeMojoModel.StringDecisionPathTracker

Field Summary

Fields

Modifier and Type	Field and Description
protected double[]	_calib_glm_beta GLM's beta used for calibrating output probabilities using Platt Scaling.
protected byte[][]	_compressed_trees Array of binary tree data, each tree being a byte[] array.
protected byte[][]	_compressed_trees_aux Array of auxiliary binary tree data, each being a byte[] array.
protected double	_mojo_version
protected int	_ntree_groups _ntree_groups is the number of trees requested by the user.
protected int	_ntrees_per_group

Fields inherited from class hex.genmodel.MojoModel

_balanceClasses, _category, _defaultThreshold, _modelClassDistrib, _nclasses, _nfeatures, _priorClassDistrib, _supervised, _uuid

Fields inherited from class hex.genmodel.GenModel

_domains, _names, _offsetColumn, _responseColumn

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	SharedTreeMojoModel(java.lang.String[] columns, java.lang.String[][] domains, java.lang.String responseColumn)

Method Summary

Modifier and Type	Method and Description
SharedTreeGraph	_computeGraph(int treeToPrint) Compute a graph of the forest.
boolean	calibrateClassProbabilities(double[] preds) Subclasses implement calibration of class probabilities.
static SharedTreeSubgraph	computeTreeGraph(int treeNum, java.lang.String treeName, byte[] tree, byte[] auxTreeInfo, int nclasses, java.lang.String[] names, java.lang.String[][] domains)
static java.lang.String	getDecisionPath(double leafAssignment)
java.lang.String[]	getDecisionPath(double[] row)
static <T> T	getDecisionPath(double leafAssignment, SharedTreeMojoModel.DecisionPathTracker<T> tr)
java.lang.String[]	getDecisionPathNames()
SharedTreeMojoModel.LeafNodeAssignments	getLeafNodeAssignments(double[] row)
static int	getLeafNodeId(double leafAssignment, byte[] auxTree)
int	getNTreeGroups()
int	getNTreesPerGroup()
protected void	postInit()
protected void	scoreAllTrees(double[] row, double[] preds) Score all trees and fill in the `preds` array.
void	scoreSingleTree(double[] row, int index, double[] preds) Generates a (per-class) prediction using only a single tree.
static double	scoreTree(byte[] tree, double[] row, int nclasses, boolean computeLeafAssignment, java.lang.String[][] domains) Highly efficient (critical path) tree scoring Given a tree (in the form of a byte array) and the row of input data, compute either this tree's predicted value when `computeLeafAssignment` is false, or the the decision path within the tree (but no more than 64 levels) when `computeLeafAssignment` is true.
static double	scoreTree0(byte[] tree, double[] row, int nclasses, boolean computeLeafAssignment) SET IN STONE FOR MOJO VERSION "1.00" - DO NOT CHANGE
static double	scoreTree1(byte[] tree, double[] row, int nclasses, boolean computeLeafAssignment) SET IN STONE FOR MOJO VERSION "1.10" - DO NOT CHANGE
void	scoreTreeRange(double[] row, int fromIndex, int toIndex, double[] preds) Generates (partial, per-class) predictions using only trees from a given range.
static java.lang.String	treeName(int groupIndex, int classIndex, java.lang.String[] domainValues)
abstract double[]	unifyPreds(double[] row, double offset, double[] preds) Transforms tree predictions into the final model predictions.

Methods inherited from class hex.genmodel.MojoModel

getModelCategory, getUUID, isSupervised, load, load, nclasses, nfeatures

Methods inherited from class hex.genmodel.GenModel

bitSetContains, bitSetIsInRange, convertDouble2Float, correctProbabilities, createAuxKey, GBM_rescale, getColIdx, getDomainValues, getDomainValues, getDomainValues, getHeader, getModelCategories, getNames, getNumClasses, getNumCols, getNumResponseClasses, getPrediction, getPredsSize, getPredsSize, getResponseIdx, getResponseName, GLM_identityInv, GLM_inverseInv, GLM_logInv, GLM_logitInv, GLM_ologitInv, GLM_tweedieInv, img2pixels, isAutoEncoder, isClassifier, KMeans_closest, KMeans_distance, KMeans_distance, KMeans_distances, Kmeans_preprocessData, Kmeans_preprocessData, KMeans_simplex, log_rescale, mapEnum, score0, score0, setCats, setCats, setInput, setInput

Methods inherited from class java.lang.Object

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

Field Detail

_mojo_version

protected double _mojo_version

_ntree_groups

protected int _ntree_groups

_ntree_groups is the number of trees requested by the user. For binomial case or regression this is also the total number of trees trained; however in multinomial case each requested "tree" is actually represented as a group of trees, with _ntrees_per_group trees in each group. Each of these individual trees assesses the likelihood that a given observation belongs to class A, B, C, etc. of a multiclass response.

_ntrees_per_group

protected int _ntrees_per_group

_compressed_trees

protected byte[][] _compressed_trees

Array of binary tree data, each tree being a byte[] array. The trees are logically grouped into a rectangular grid of dimensions _ntree_groups x _ntrees_per_group, however physically they are stored as 1-dimensional list, and an [i, j] logical tree is mapped to the index treeIndex(int, int).

_compressed_trees_aux

protected byte[][] _compressed_trees_aux

Array of auxiliary binary tree data, each being a byte[] array.

_calib_glm_beta

protected double[] _calib_glm_beta

GLM's beta used for calibrating output probabilities using Platt Scaling.

Constructor Detail

SharedTreeMojoModel

protected SharedTreeMojoModel(java.lang.String[] columns,
                              java.lang.String[][] domains,
                              java.lang.String responseColumn)

Method Detail

postInit

protected void postInit()

getNTreeGroups

public final int getNTreeGroups()

getNTreesPerGroup

public final int getNTreesPerGroup()

scoreTree

public static double scoreTree(byte[] tree,
                               double[] row,
                               int nclasses,
                               boolean computeLeafAssignment,
                               java.lang.String[][] domains)

Highly efficient (critical path) tree scoring Given a tree (in the form of a byte array) and the row of input data, compute either this tree's predicted value when `computeLeafAssignment` is false, or the the decision path within the tree (but no more than 64 levels) when `computeLeafAssignment` is true. Note: this function is also used from the `hex.tree.CompressedTree` class in `h2o-algos` project.

getDecisionPath

public static <T> T getDecisionPath(double leafAssignment,
                                    SharedTreeMojoModel.DecisionPathTracker<T> tr)

getDecisionPath

public static java.lang.String getDecisionPath(double leafAssignment)

getLeafNodeId

public static int getLeafNodeId(double leafAssignment,
                                byte[] auxTree)

_computeGraph

public SharedTreeGraph _computeGraph(int treeToPrint)

Compute a graph of the forest.

Returns:

A graph of the forest.

computeTreeGraph

public static SharedTreeSubgraph computeTreeGraph(int treeNum,
                                                  java.lang.String treeName,
                                                  byte[] tree,
                                                  byte[] auxTreeInfo,
                                                  int nclasses,
                                                  java.lang.String[] names,
                                                  java.lang.String[][] domains)

treeName

public static java.lang.String treeName(int groupIndex,
                                        int classIndex,
                                        java.lang.String[] domainValues)

scoreAllTrees

protected void scoreAllTrees(double[] row,
                             double[] preds)

Score all trees and fill in the `preds` array.

unifyPreds

public abstract double[] unifyPreds(double[] row,
                                    double offset,
                                    double[] preds)

Transforms tree predictions into the final model predictions. For classification: converts tree preds into probability distribution and picks predicted class. For regression: projects tree prediction from link-space into the original space.

Parameters:

row - input row.

offset - offset.

preds - final output, same structure as of GenModel.score0(double[], double[]).

Returns:

preds array.

scoreSingleTree

public final void scoreSingleTree(double[] row,
                                  int index,
                                  double[] preds)

Generates a (per-class) prediction using only a single tree.

Parameters:

row - input row

index - index of the tree (0..N-1)

preds - array of partial predictions.

scoreTreeRange

public final void scoreTreeRange(double[] row,
                                 int fromIndex,
                                 int toIndex,
                                 double[] preds)

Generates (partial, per-class) predictions using only trees from a given range.

Parameters:

row - input row

fromIndex - low endpoint (inclusive) of the tree range

toIndex - high endpoint (exclusive) of the tree range

preds - array of partial predictions. To get final predictions pass the result to unifyPreds(double[], double, double[]).

getDecisionPathNames

public java.lang.String[] getDecisionPathNames()

getLeafNodeAssignments

public SharedTreeMojoModel.LeafNodeAssignments getLeafNodeAssignments(double[] row)

getDecisionPath

public java.lang.String[] getDecisionPath(double[] row)

scoreTree0

public static double scoreTree0(byte[] tree,
                                double[] row,
                                int nclasses,
                                boolean computeLeafAssignment)

SET IN STONE FOR MOJO VERSION "1.00" - DO NOT CHANGE

Parameters:

tree -

row -

nclasses -

computeLeafAssignment -

Returns:

scoreTree1

public static double scoreTree1(byte[] tree,
                                double[] row,
                                int nclasses,
                                boolean computeLeafAssignment)

SET IN STONE FOR MOJO VERSION "1.10" - DO NOT CHANGE

Parameters:

tree -

row -

nclasses -

computeLeafAssignment -

Returns:

calibrateClassProbabilities

public boolean calibrateClassProbabilities(double[] preds)

Description copied from class: GenModel

Subclasses implement calibration of class probabilities. The input is array of predictions returned by the scoring function (score0). Supports classification models that were trained with calibration enabled. Original probabilities in the predictions array are overwritten by their corresponding calibrated counterparts. Return false if model doesn't support calibration.

Overrides:

calibrateClassProbabilities in class GenModel