hex

Class ModelMetrics.MetricBuilder<T extends ModelMetrics.MetricBuilder<T>>

java.lang.Object

water.Iced<T>

hex.ModelMetrics.MetricBuilder<T>

All Implemented Interfaces:

java.io.Externalizable, java.io.Serializable, java.lang.Cloneable, Freezable<T>

Direct Known Subclasses:

ModelMetricsSupervised.MetricBuilderSupervised, ModelMetricsUnsupervised.MetricBuilderUnsupervised

Enclosing class:

ModelMetrics

public abstract static class ModelMetrics.MetricBuilder<T extends ModelMetrics.MetricBuilder<T>>
extends Iced<T>

Class used to compute AUCs, CMs & HRs "on the fly" during other passes over Big Data. This class is intended to be embedded in other MRTask objects. The perRow method is called once-per-scored-row, and the reduce method called once per MRTask.reduce, and the <init> called once per MRTask.map.

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
long	_count
CustomMetric	_customMetric
double	_sumsqe
double	_wcount
double[]	_work
double	_wY
double	_wYY

Constructor Summary

Constructors

Constructor and Description
MetricBuilder()

Method Summary

Modifier and Type	Method and Description
void	cachePrediction(double[] cdist, Chunk[] chks, int row, int cacheChunkIdx, Model m)
abstract ModelMetrics	makeModelMetrics(Model m, Frame f, Frame adaptedFrame, Frame preds) Having computed a MetricBuilder, this method fills in a ModelMetrics
Frame	makePredictionCache(Model m, Vec response)
double[]	perRow(double[] ds, float[] yact, double weight, double offset, Model m)
abstract double[]	perRow(double[] ds, float[] yact, Model m)
void	postGlobal()
void	postGlobal(CustomMetric customMetric)
void	reduce(java.lang.Object mb)
void	reduce(T mb)
void	reduceForCV(T mb)
void	setCustomMetric(CustomMetric customMetric) Set value of custom metric.
double	weightedSigma()

Methods inherited from class water.Iced

asBytes, clone, copyOver, frozenType, read, readExternal, readJSON, reloadFromBytes, toJsonBytes, toJsonString, write, writeExternal, writeJSON

Methods inherited from class java.lang.Object

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

Field Detail

_work

public transient double[] _work

_sumsqe

public double _sumsqe

_count

public long _count

_wcount

public double _wcount

_wY

public double _wY

_wYY

public double _wYY

_customMetric

public CustomMetric _customMetric

Constructor Detail

MetricBuilder

public MetricBuilder()

Method Detail

weightedSigma

public double weightedSigma()

perRow

public abstract double[] perRow(double[] ds,
                                float[] yact,
                                Model m)

perRow

public double[] perRow(double[] ds,
                       float[] yact,
                       double weight,
                       double offset,
                       Model m)

reduce

public void reduce(T mb)

reduce

public void reduce(java.lang.Object mb)

reduceForCV

public void reduceForCV(T mb)

postGlobal

public void postGlobal()

postGlobal

public void postGlobal(CustomMetric customMetric)

makeModelMetrics

public abstract ModelMetrics makeModelMetrics(Model m,
                                              Frame f,
                                              Frame adaptedFrame,
                                              Frame preds)

Having computed a MetricBuilder, this method fills in a ModelMetrics

Parameters:

m - Model

f - Scored Frame

adaptedFrame - Adapted Frame

preds - Predictions of m on f (optional) @return Filled Model Metrics object

setCustomMetric

public void setCustomMetric(CustomMetric customMetric)

Set value of custom metric.

Parameters:

customMetric - computed custom metric outside of this default builder

makePredictionCache

public Frame makePredictionCache(Model m,
                                 Vec response)

cachePrediction

public void cachePrediction(double[] cdist,
                            Chunk[] chks,
                            int row,
                            int cacheChunkIdx,
                            Model m)