hex.rf
Class RFModel

java.lang.Object
  water.Iced
      water.OldModel
          hex.rf.RFModel

All Implemented Interfaces:

java.lang.Cloneable, Freezable, Job.Progress

public class RFModel
extends OldModel
implements Job.Progress

A model is an ensemble of trees that can be serialized and that can be used to classify data.

Field Summary

int	_features Number of features these trees are built for
Key[][]	_localForests Local forests produced by nodes
int[]	_nodesSplitFeatures Number of computed split features per node - number of split features can differ for each node.
Key[][]	_remoteChunksKeys Remote chunks' keys used by individual nodes
float	_sample Sampling rate used when building trees.
Sampling.Strategy	_samplingStrategy Sampling strategy used for model
int	_splitFeatures Number of split features defined by user.
float[]	_strataSamples Strata sampling rate used for local-node strata-sampling
long	_time Total time in seconds to produce model
Key[]	_tkeys All the trees in the model
int	_totalTrees Number of keys the model expects to be built for it
byte[][]	_trees
static java.lang.String	KEY_PREFIX

Fields inherited from class water.OldModel

_dataKey, _selfKey, _va, DOC_FIELDS

Constructor Summary

RFModel()
Empty constructor for deserialization

RFModel(Key selfKey, int[] cols, Key dataKey, Key[] tkeys, int features, Sampling.Strategy samplingStrategy, float sample, float[] strataSamples, int splitFeatures, int totalTrees)
A RandomForest Model

RFModel(Key selfKey, java.lang.String[] colNames, java.lang.String[] classNames, Key[] tkeys, int features, float sample)

Method Summary

int	classes()
short	classify(int[] votes, double[] classWt, java.util.Random rand)
short	classify(ValueArray data, AutoBuffer chunk, int row, int[] modelDataMap, int[] votes, double[] classWt, java.util.Random rand)
short	classify0(int tree_id, ValueArray data, AutoBuffer chunk, int row, int[] modelDataMap, short badrow) Classify a row according to one particular tree.
void	deleteKeys() Bad name, I know.
Counter	depth()
void	find_leaves_depth() Internal computation of depth and number of leaves.
long	getTreeSeed(int i) Return the random seed used to sample this tree.
Counter	leaves()
static RFModel	make(RFModel old, Key tkey, int nodeIdx)
static Key	makeKey()
java.lang.String	name(int atree)
float	progress()
protected double	score0(double[] data) Single row scoring, on properly ordered data
protected double	score0(ValueArray data, AutoBuffer ab, int row_in_chunk) Bulk scoring API, on a compatible ValueArray (when pushed throw the mapping)
protected double	score0(ValueArray data, int row) Single row scoring, on a compatible ValueArray (when pushed throw the mapping)
int	size()
com.google.gson.JsonObject	toJson()
byte[]	tree(int tree_id) Return the bits for a particular tree
int	treeCount() The number of trees in this model.

Methods inherited from class water.OldModel

adapt, adapt, columnFilter, columnMapping, delete, fromJson, identityMap, isCompatible, isCompatible, isCompatible, response, responseName, score, score

Methods inherited from class water.Iced

clone, frozenType, init, newInstance, read, toDocField, write, writeJSON, writeJSONFields

Methods inherited from class java.lang.Object

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

Field Detail

_features

public int _features

Number of features these trees are built for

_samplingStrategy

public Sampling.Strategy _samplingStrategy

Sampling strategy used for model

_sample

public float _sample

Sampling rate used when building trees.

_strataSamples

public float[] _strataSamples

Strata sampling rate used for local-node strata-sampling

_splitFeatures

public int _splitFeatures

Number of split features defined by user.

_nodesSplitFeatures

public int[] _nodesSplitFeatures

Number of computed split features per node - number of split features can differ for each node. However, such difference would point to a problem with data distribution.

_totalTrees

public int _totalTrees

Number of keys the model expects to be built for it

_tkeys

public Key[] _tkeys

All the trees in the model

_localForests

public Key[][] _localForests

Local forests produced by nodes

_remoteChunksKeys

public Key[][] _remoteChunksKeys

Remote chunks' keys used by individual nodes

_time

public long _time

Total time in seconds to produce model

_trees

public transient byte[][] _trees

KEY_PREFIX

public static final java.lang.String KEY_PREFIX

See Also:

Constant Field Values

Constructor Detail

RFModel

public RFModel(Key selfKey,
               int[] cols,
               Key dataKey,
               Key[] tkeys,
               int features,
               Sampling.Strategy samplingStrategy,
               float sample,
               float[] strataSamples,
               int splitFeatures,
               int totalTrees)

A RandomForest Model

Parameters:

treeskey - a key of keys of trees

classes - the number of response classes

data - the dataset

RFModel

public RFModel(Key selfKey,
               java.lang.String[] colNames,
               java.lang.String[] classNames,
               Key[] tkeys,
               int features,
               float sample)

RFModel

public RFModel()

Empty constructor for deserialization

Method Detail

make

public static RFModel make(RFModel old,
                           Key tkey,
                           int nodeIdx)

makeKey

public static final Key makeKey()

treeCount

public int treeCount()

The number of trees in this model.

size

public int size()

classes

public int classes()

progress

public float progress()

Specified by:

progress in interface Job.Progress

name

public java.lang.String name(int atree)

tree

public byte[] tree(int tree_id)

Return the bits for a particular tree

deleteKeys

public void deleteKeys()

Bad name, I know. But free all internal tree keys.

classify0

public short classify0(int tree_id,
                       ValueArray data,
                       AutoBuffer chunk,
                       int row,
                       int[] modelDataMap,
                       short badrow)

Classify a row according to one particular tree.

Parameters:

tree_id - the number of the tree to use

chunk - the chunk we are using

row - the row number in the chunk

modelDataMap - mapping from model/tree columns to data columns

Returns:

the predicted response class, or class+1 for broken rows

classify

public short classify(ValueArray data,
                      AutoBuffer chunk,
                      int row,
                      int[] modelDataMap,
                      int[] votes,
                      double[] classWt,
                      java.util.Random rand)

classify

public short classify(int[] votes,
                      double[] classWt,
                      java.util.Random rand)

find_leaves_depth

public void find_leaves_depth()

Internal computation of depth and number of leaves.

leaves

public Counter leaves()

depth

public Counter depth()

getTreeSeed

public long getTreeSeed(int i)

Return the random seed used to sample this tree.

score0

protected double score0(double[] data)

Single row scoring, on properly ordered data

Specified by:

score0 in class OldModel

score0

protected double score0(ValueArray data,
                        int row)

Single row scoring, on a compatible ValueArray (when pushed throw the mapping)

Overrides:

score0 in class OldModel

score0

protected double score0(ValueArray data,
                        AutoBuffer ab,
                        int row_in_chunk)

Bulk scoring API, on a compatible ValueArray (when pushed throw the mapping)

Overrides:

score0 in class OldModel

toJson

public com.google.gson.JsonObject toJson()

Overrides:

toJson in class OldModel