hex.kmeans

Class KMeansGrid

java.lang.Object

water.Iced<T>

water.Keyed<T>

water.Lockable<G>

hex.Grid<KMeansGrid>

hex.kmeans.KMeansGrid

All Implemented Interfaces:

java.io.Externalizable, java.io.Serializable, java.lang.Cloneable, water.Freezable<KMeansGrid>

public class KMeansGrid
extends hex.Grid<KMeansGrid>

A Grid of Models Used to explore Model hyper-parameter space. Lazily filled in, this object represents the potentially infinite variety of hyperparameters of a given model & dataset. One subclass per kind of Model, e.g. KMeans or GLM or GBM or DL. The Grid tracks Models and their hyperparameters, and will allow discovery of existing Models by hyperparameter, or building Models on demand by hyperparameter. The Grid can manage a (simplistic) hyperparameter search space. Hyperparameter values are limited to doubles in the API, but can be anything the subclass Grid desires internally. E.g. the Grid for KMeans will convert the initial center selection Enum to and from a simple integer value internally.

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from class hex.Grid

hex.Grid.GridSearch

Field Summary

Fields

Modifier and Type	Field and Description
static java.lang.String	MODEL_NAME

Fields inherited from class hex.Grid

_fr

Fields inherited from class water.Lockable

_lockers

Fields inherited from class water.Keyed

_key

Method Summary

Methods

Modifier and Type	Method and Description
protected long	checksum_impl()
static KMeansGrid	get(water.fvec.Frame fr)
protected KMeans	getBuilder(double[] hypers)
double[]	getHypers(hex.Model.Parameters parms)
protected double[]	hyperDefaults()
protected java.lang.String[]	hyperNames()
protected java.lang.String	modelName()
protected double	suggestedNextHyperValue(int h, hex.Model m, double[] hyperLimits) Ask the Grid for a suggested next hyperparameter value, given an existing Model as a starting point and the complete set of hyperparameter limits.

Methods inherited from class hex.Grid

keyName, model, model, remove_impl, startGridSearch, trainingFrame

Methods inherited from class water.Lockable

delete_and_lock, delete, delete, delete, read_lock, read_lock, unlock_all, unlock, update, write_lock

Methods inherited from class water.Keyed

checksum, remove, remove, remove, remove

Methods inherited from class water.Iced

clone, frozenType, read_impl, read, readExternal, readJSON_impl, readJSON, toJsonString, write_impl, write, writeExternal, writeHTML_impl, writeHTML, writeJSON_impl, writeJSON

Methods inherited from class java.lang.Object

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

Field Detail

MODEL_NAME

public static final java.lang.String MODEL_NAME

See Also:

Constant Field Values

Method Detail

modelName

protected java.lang.String modelName()

Specified by:

modelName in class hex.Grid<KMeansGrid>

Returns:

Model name

hyperNames

protected java.lang.String[] hyperNames()

Specified by:

hyperNames in class hex.Grid<KMeansGrid>

Returns:

hyperparameter names corresponding to a Model.Parameter field names

hyperDefaults

protected double[] hyperDefaults()

Specified by:

hyperDefaults in class hex.Grid<KMeansGrid>

Returns:

hyperparameter defaults, aligned with the field names

suggestedNextHyperValue

protected double suggestedNextHyperValue(int h,
                             hex.Model m,
                             double[] hyperLimits)

Ask the Grid for a suggested next hyperparameter value, given an existing Model as a starting point and the complete set of hyperparameter limits. Returning a NaN signals there is no next suggestion, which is reasonable if the obvious "next" value does not exist (e.g. exhausted all possibilities of an enum). It is OK if a Model for the suggested value already exists; this will be checked before building any model.

Specified by:

suggestedNextHyperValue in class hex.Grid<KMeansGrid>

Parameters:

h - The h-th hyperparameter

m - A model to act as a starting point

hyperLimits - Upper bounds for this search

Returns:

Suggested next value for hyperparameter h or NaN if no next value

getBuilder

protected KMeans getBuilder(double[] hypers)

Specified by:

getBuilder in class hex.Grid<KMeansGrid>

Parameters:

hypers - A set of hyper parameter values

Returns:

A ModelBuilder, blindly filled with parameters. Assumed to be cheap; used to check hyperparameter sanity or make models

getHypers

public double[] getHypers(hex.Model.Parameters parms)

Specified by:

getHypers in class hex.Grid<KMeansGrid>

Parameters:

parms - Model parameters

Returns:

Gridable parameters pulled out of the parms

get

public static KMeansGrid get(water.fvec.Frame fr)

checksum_impl

protected long checksum_impl()

Specified by:

checksum_impl in class water.Keyed<KMeansGrid>