hex.glm

Class GLM

java.lang.Object

water.Iced

hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

hex.glm.GLM

All Implemented Interfaces:

java.io.Externalizable, java.io.Serializable, java.lang.Cloneable, water.Freezable

public class GLM
extends hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

Created by tomasnykodym on 8/27/14. Generalized linear model implementation.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	GLM.BetaConstraint
static class	GLM.BetaInfo
class	GLM.GLMDriver Main loop of the glm algo.
static class	GLM.GLMGradientInfo
static class	GLM.GLMGradientSolver Gradient and line search computation for L_BFGS and also L_BFGS solver wrapper (for ADMM)
static class	GLM.GramSolver Created by tomasnykodym on 3/30/15.
static class	GLM.PlugValuesImputer
static class	GLM.ProximalGradientInfo
static class	GLM.ProximalGradientSolver Simple wrapper around ginfo computation, adding proximal penalty

Nested classes/interfaces inherited from class hex.ModelBuilder

hex.ModelBuilder.BuilderVisibility, hex.ModelBuilder.Driver, hex.ModelBuilder.FilterCols, hex.ModelBuilder.ModelTrainingCoordinator, hex.ModelBuilder.ValidationMessage

Field Summary

Fields

Modifier and Type	Field and Description
GLM.BetaInfo	_betaInfo
int[][]	_gamColIndices
java.lang.String[][]	_gamColnames
java.lang.String	_generatedWeights
double[][][]	_penaltyMatrix
static int	SCORING_INTERVAL_MSEC
protected static long	WORK_TOTAL

Fields inherited from class hex.ModelBuilder

_coordinator, _desc, _eventPublisher, _fold, _input_parms, _job, _messages, _nclass, _offset, _orig_projection_array, _origDomains, _origNames, _origTrain, _parms, _priorClassDist, _removedCols, _response, _result, _startUpOnceModelBuilder, _train, _treatment, _valid, _vresponse, _weights

Constructor Summary

Constructors

Constructor and Description
GLM(boolean startup_once)
GLM(GLMModel.GLMParameters parms)
GLM(GLMModel.GLMParameters parms, double[][][] penaltyMatrix, java.lang.String[][] gamColnames) This constructor is only called by GAM when it is trying to build a GAM model using GLM.
GLM(GLMModel.GLMParameters parms, water.Key dest)

Method Summary

Modifier and Type	Method and Description
hex.ModelCategory[]	can_build()
protected void	checkMemoryFootPrint(DataInfo activeData)
double[]	COD_solve(ComputationState.GramXY gram, double alpha, double lambda)
void	computeCrossValidation() GLM implementation of N-fold cross-validation.
protected boolean	computePriorClassDistribution()
void	cv_computeAndSetOptimalParameters(hex.ModelBuilder[] cvModelBuilders) If run with lambda search, we need to take extra action performed after cross-val models are built.
static int[]	findIterIndexAcrossFolds(java.util.List<java.lang.Integer>[] cvModelIters, int fold0Index) This method is used to locate common iteration indices across all folds.
double[]	genInitBeta()
boolean	haveMojo()
boolean	havePojo()
void	init(boolean expensive)
boolean	isSupervised()
int	nclasses()
protected static double	sparseOffset(double[] beta, DataInfo dinfo)
protected GLM.GLMDriver	trainModelImpl()

Methods inherited from class hex.ModelBuilder

algoName, algos, builderVisibility, canLearnFromNAs, checkCustomMetricForEarlyStopping, checkDistributions, checkEarlyStoppingReproducibility, checkMemoryFootPrint_impl, checkMemoryFootPrint, checkResponseVariable, clearInitState, clearValidationErrors, cv_buildModels, cv_canBuildMainModelInParallel, cv_initStoppingParameters, cv_mainModelScores, cv_makeAggregateModelMetrics, cv_scoreCVModels, cv_updateOptimalParameters, defaultKey, desiredChunks, dest, error_count, error, get, getMessagesByFieldAndSeverity, getName, getSysProperty, getToEigenVec, hasFoldCol, hasOffsetCol, hasTreatmentCol, hasWeightCol, hide, ignoreBadColumns, ignoreConstColumns, ignoreInvalidColumns, ignoreStringColumns, ignoreUuidColumns, info, init_adaptFrameToTrain, init_getNClass, initWorkspace, isClassifier, isResponseOptional, isStopped, javaName, logMe, make, make, make, makeCVMetrics, makeCVModelBuilder, makeParameters, makePojoWriter, message, nFoldCV, nFoldWork, nModelsInParallel, nModelsInParallel, nModelsInParallel, numSpecialCols, paramName, raiseReproducibilityWarning, rebalance, remainingTimeSecs, response, schemaDirectory, separateFeatureVecs, setMaxRuntimeSecsForMainModel, setTrain, setValid, shouldReorder, smallDataSize, stop_requested, timeout, train, trainModel, trainModel, trainModelNested, trainModelNested, trainModelOnH2ONode, valid, validateBinaryResponse, validateStoppingMetric, validationErrors, validationWarnings, vresponse, warn

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

SCORING_INTERVAL_MSEC

public static final int SCORING_INTERVAL_MSEC

See Also:

Constant Field Values

_generatedWeights

public java.lang.String _generatedWeights

_penaltyMatrix

public double[][][] _penaltyMatrix

_gamColnames

public java.lang.String[][] _gamColnames

_gamColIndices

public int[][] _gamColIndices

_betaInfo

public GLM.BetaInfo _betaInfo

WORK_TOTAL

protected static final long WORK_TOTAL

See Also:

Constant Field Values

Constructor Detail

GLM

public GLM(boolean startup_once)

GLM

public GLM(GLMModel.GLMParameters parms)

GLM

public GLM(GLMModel.GLMParameters parms,
           double[][][] penaltyMatrix,
           java.lang.String[][] gamColnames)

This constructor is only called by GAM when it is trying to build a GAM model using GLM. Internal function, DO NOT USE.

GLM

public GLM(GLMModel.GLMParameters parms,
           water.Key dest)

Method Detail

isSupervised

public boolean isSupervised()

Specified by:

isSupervised in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

can_build

public hex.ModelCategory[] can_build()

Specified by:

can_build in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

havePojo

public boolean havePojo()

Overrides:

havePojo in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

haveMojo

public boolean haveMojo()

Overrides:

haveMojo in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

computeCrossValidation

public void computeCrossValidation()

GLM implementation of N-fold cross-validation. We need to compute the sequence of lambdas for the main model so the folds share the same lambdas. We also want to set the _cv flag so that the dependent jobs know they're being run withing CV (so e.g. they do not unlock the models in the end)

Overrides:

computeCrossValidation in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

cv_computeAndSetOptimalParameters

public void cv_computeAndSetOptimalParameters(hex.ModelBuilder[] cvModelBuilders)

If run with lambda search, we need to take extra action performed after cross-val models are built. Each of the folds have been computed with ots own private validation dataset and it performed early stopping based on it. => We need to: 1. compute cross-validated lambda estimate 2. set the lambda estimate to all n-folds models (might require extra model fitting if the particular model stopped too early!) 3. compute cross-validated scoring history (cross-validated deviance standard error per lambda) 4. unlock the n-folds models (they are changed here, so the unlocking happens here)

Overrides:

cv_computeAndSetOptimalParameters in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

findIterIndexAcrossFolds

public static int[] findIterIndexAcrossFolds(java.util.List<java.lang.Integer>[] cvModelIters,
                                             int fold0Index)

This method is used to locate common iteration indices across all folds. Since scoring history is scored by user specifying the scoring interval and by a time interval determined by system computation speed, there can be scoring history with different scoring indices across folds. This method is used to find the commen iteration indices and return it.

Parameters:

cvModelIters: - store model keys from models generated by cross validation.

fold0Index: - iteration index of fold 0 model

checkMemoryFootPrint

protected void checkMemoryFootPrint(DataInfo activeData)

nclasses

public int nclasses()

Overrides:

nclasses in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

computePriorClassDistribution

protected boolean computePriorClassDistribution()

Overrides:

computePriorClassDistribution in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

init

public void init(boolean expensive)

Overrides:

init in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

genInitBeta

public double[] genInitBeta()

trainModelImpl

protected GLM.GLMDriver trainModelImpl()

Specified by:

trainModelImpl in class hex.ModelBuilder<GLMModel,GLMModel.GLMParameters,GLMModel.GLMOutput>

COD_solve

public double[] COD_solve(ComputationState.GramXY gram,
                          double alpha,
                          double lambda)

sparseOffset

protected static double sparseOffset(double[] beta,
                                     DataInfo dinfo)