hex.rf

Class ConfusionTask

java.lang.Object

jsr166y.ForkJoinTask<java.lang.Void>

jsr166y.CountedCompleter

water.H2O.H2OCountedCompleter

water.DTask<T>

water.DRemoteTask<T>

water.MRTask

hex.rf.ConfusionTask

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable, java.util.concurrent.Future<java.lang.Void>, jsr166y.ForkJoinPool.ManagedBlocker, Freezable

public class ConfusionTask
extends MRTask

Confusion Matrix. Incrementally computes a Confusion Matrix for a forest of Trees, vs a given input dataset. The set of Trees can grow over time. Each request from the Confusion compute on any new trees (if any), and report a matrix. Cheap if all trees already computed.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	ConfusionTask.CMFinal
static class	ConfusionTask.CMJob

Field Summary

Fields

Modifier and Type	Field and Description
ValueArray	_data LOCAL: The dataset to validate
int	_DATA_N LOCAL: Number of response classes in data
int	_MODEL_N LOCAL: Number of response classes in model
int	_N LOCAL: Number of response classes = Max(responses in model, responses in test data)

Fields inherited from class water.MRTask

_hi, _lo, _runSingleThreaded

Fields inherited from class water.DRemoteTask

_fs, _is_local, _keys, _top_level

Fields inherited from class water.DTask

_cls, _eFromNode, _exception, _fname, _lineNum, _msg, _mth

Constructor Summary

Constructors

Constructor and Description
ConfusionTask() Constructor for use by the serializers

Method Summary

Methods

Modifier and Type	Method and Description
static int	alignEnumDomains(java.lang.String[] modelDomain, java.lang.String[] dataDomain, int[] modelMapping, int[] dataMapping) Merge model and data predictor domain to produce domain for CM.
int	dimension() Return number of classes - in fact dimension of CM.
java.lang.String[]	domain() Compute confusion matrix domain based on model and data key.
static java.lang.String[]	domain(int N, ValueArray.Column modelCol, ValueArray.Column dataCol, int[] modelEnumMapping, int[] dataEnumMapping)
static java.lang.String[]	domain(ValueArray.Column modelCol, ValueArray.Column dataCol)
void	init() Once-per-remote invocation init.
Key	keyForCM()
static Key	keyForCM(Key modelKey, int msize, Key datakey, int classcol, boolean computeOOB)
static Job	make(RFModel model, int modelSize, Key datakey, int classcol, double[] classWt, boolean computeOOB)
static Job	make(RFModel model, Key datakey, int classcol, double[] classWt, boolean computeOOB) Apply a model to a dataset to produce a Confusion Matrix.
void	map(Key chunkKey) A classic Map/Reduce style incremental computation of the confusion matrix on a chunk of data.
void	reduce(DRemoteTask drt) Reduction combines the confusion matrices.
static void	remove(RFModel model, Key datakey, int classcol, boolean computeOOB)

Methods inherited from class water.MRTask

hi, lcompute, lo, lonCompletion, memOverheadPerChunk, onExceptionalCompletion

Methods inherited from class water.DRemoteTask

alsoBlockFor, alsoBlockFor, block, clone, compute2, dfork, getFutures, invoke, invokeOnAllNodes, isReleasable, keys, onCompletion, postGlobal, reduceAlsoBlock

Methods inherited from class water.DTask

copyOver, dinvoke, frozenType, getDException, hasException, logVerbose, newInstance, onAck, onAckAck, read, setException, toDocField, write, writeJSONFields

Methods inherited from class water.H2O.H2OCountedCompleter

compute, priority

Methods inherited from class jsr166y.CountedCompleter

addToPendingCount, compareAndSetPendingCount, complete, exec, getCompleter, getPendingCount, getRawResult, setCompleter, setPendingCount, setRawResult, tryComplete

Methods inherited from class jsr166y.ForkJoinTask

adapt, adapt, adapt, cancel, compareAndSetForkJoinTaskTag, completeExceptionally, fork, get, get, getException, getForkJoinTaskTag, getPool, getQueuedTaskCount, getSurplusQueuedTaskCount, helpQuiesce, inForkJoinPool, invoke, invokeAll, invokeAll, invokeAll, isCancelled, isCompletedAbnormally, isCompletedNormally, isDone, join, peekNextLocalTask, pollNextLocalTask, pollTask, quietlyComplete, quietlyInvoke, quietlyJoin, reinitialize, setForkJoinTaskTag, tryUnfork

Methods inherited from class java.lang.Object

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

Field Detail

_data

public transient ValueArray _data

LOCAL: The dataset to validate

_N

public transient int _N

LOCAL: Number of response classes = Max(responses in model, responses in test data)

_MODEL_N

public transient int _MODEL_N

LOCAL: Number of response classes in model

_DATA_N

public transient int _DATA_N

LOCAL: Number of response classes in data

Constructor Detail

ConfusionTask

public ConfusionTask()

Constructor for use by the serializers

Method Detail

keyForCM

public Key keyForCM()

keyForCM

public static Key keyForCM(Key modelKey,
           int msize,
           Key datakey,
           int classcol,
           boolean computeOOB)

remove

public static void remove(RFModel model,
          Key datakey,
          int classcol,
          boolean computeOOB)

make

public static Job make(RFModel model,
       Key datakey,
       int classcol,
       double[] classWt,
       boolean computeOOB)

Apply a model to a dataset to produce a Confusion Matrix. To support incremental and repeated model application, hash the model and data and look for that Key to already exist, returning a prior CM if one is available.

make

public static Job make(RFModel model,
       int modelSize,
       Key datakey,
       int classcol,
       double[] classWt,
       boolean computeOOB)

init

public void init()

Once-per-remote invocation init. The standard M/R framework will endlessly clone the original object "for free" (well, for very low cost), but the wire-line format does not send over things we can compute locally. So compute locally, once, some things we want in all cloned instances.

Overrides:

init in class MRTask

map

public void map(Key chunkKey)

A classic Map/Reduce style incremental computation of the confusion matrix on a chunk of data.

Specified by:

map in class MRTask

reduce

public void reduce(DRemoteTask drt)

Reduction combines the confusion matrices.

Specified by:

reduce in class DRemoteTask

alignEnumDomains

public static int alignEnumDomains(java.lang.String[] modelDomain,
                   java.lang.String[] dataDomain,
                   int[] modelMapping,
                   int[] dataMapping)

Merge model and data predictor domain to produce domain for CM. The domain is expected to be ordered and containing unique values.

domain

public static java.lang.String[] domain(ValueArray.Column modelCol,
                        ValueArray.Column dataCol)

domain

public static java.lang.String[] domain(int N,
                        ValueArray.Column modelCol,
                        ValueArray.Column dataCol,
                        int[] modelEnumMapping,
                        int[] dataEnumMapping)

domain

public java.lang.String[] domain()

Compute confusion matrix domain based on model and data key.

dimension

public final int dimension()

Return number of classes - in fact dimension of CM.