hex

Class Grid<MP extends Model.Parameters,G extends Grid<MP,G>>

java.lang.Object

water.Iced<T>

water.Keyed<T>

water.Lockable<G>

hex.Grid<MP,G>

Type Parameters:

MP - type of model build parameters

G - self-type of actual Grid implementation

All Implemented Interfaces:

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

public class Grid<MP extends Model.Parameters,G extends Grid<MP,G>>
extends Lockable<G>

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. The external Grid API uses a HashMap to describe a set of hyperparameter values, where the String is a valid field name in the corresponding Model.Parameter, and the Object is the field value (boxed as needed). The Grid implementation treats all hyperparameters as double values internally, indexed by a simple number. A complete set of hyper parameters is thus a double[], and a set of search parameters a double[][]. The subclasses of Grid will need to convert between these two formats. E.g. KMeansGrid will convert the initial center selection field "_init" Enum to and from a simple double value internally.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	Grid.GridSearch

Nested classes/interfaces inherited from class water.Keyed

Keyed.BinarySerializer<X extends Keyed>

Field Summary

Fields

Modifier and Type	Field and Description
protected Frame	_fr

Fields inherited from class water.Lockable

_lockers

Fields inherited from class water.Keyed

_key, EMPTY_KEY_LIST

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	Grid(Key key, Frame fr)

Method Summary

Methods

Modifier and Type	Method and Description
protected MP	applyHypers(MP params, double[] hypers) Apply given hyper parameters on default parameter holder.
protected long	checksum_impl() High-quality 64-bit checksum of the content of the object.
protected ModelBuilder	createBuilder(MP params) Create a new model builder for given parameters.
protected ModelBuilder	getBuilder(MP params, double[] hypers) Get a new model builder for given default parameters and hyper parameters.
double[]	getHypers(MP parms)
protected double[]	hyperDefaults()
protected java.lang.String[]	hyperNames()
protected static Key	keyName(java.lang.String modelName, Frame fr)
Key<Model>	model(double[] hypers)
Key<Model>	model(java.util.Map<java.lang.String,java.lang.Object> hypers)
protected java.lang.String	modelName()
protected Futures	remove_impl(Futures fs) Override to remove subparts, but not self, of composite Keyed objects.
Grid.GridSearch	startGridSearch(MP params, java.util.Map<java.lang.String,java.lang.Object[]> hyperSearch)
protected double	suggestedNextHyperValue(int h, 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.
Frame	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, getBinarySerializer, getPublishedKeys, 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

_fr

protected final Frame _fr

Constructor Detail

Grid

protected Grid(Key key,
    Frame fr)

Method Detail

modelName

protected java.lang.String modelName()

Returns:

Model name

hyperNames

protected java.lang.String[] hyperNames()

Returns:

hyperparameter names corresponding to a Model.Parameter field names

hyperDefaults

protected double[] hyperDefaults()

Returns:

hyperparameter defaults, aligned with the field names

suggestedNextHyperValue

protected double suggestedNextHyperValue(int h,
                             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.

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

trainingFrame

public Frame trainingFrame()

Returns:

The data frame used to train all these models. All models are trained on the same data frame, but might be validated on multiple different frames.

keyName

protected static Key keyName(java.lang.String modelName,
          Frame fr)

Returns:

Factory method to return the grid for a particular modeling class and frame.

model

public Key<Model> model(double[] hypers)

Parameters:

hypers - A set of hyper parameter values

Returns:

A model run with these parameters, or null if the model does not exist.

model

public Key<Model> model(java.util.Map<java.lang.String,java.lang.Object> hypers)

getBuilder

protected final ModelBuilder getBuilder(MP params,
                      double[] hypers)

Get a new model builder for given default parameters and hyper parameters.

Parameters:

params - default parameters

hypers - A set of hyper parameter values

Returns:

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

createBuilder

protected ModelBuilder createBuilder(MP params)

Create a new model builder for given parameters.

Parameters:

params - model builder parameters, it is private copy provided for a model builder

Returns:

applyHypers

protected MP applyHypers(MP params,
             double[] hypers)

Apply given hyper parameters on default parameter holder.

Parameters:

params - default parameters

hypers - hyper parameters to apply

Returns:

modified params object with configured hyper parameters

getHypers

public double[] getHypers(MP parms)

Parameters:

parms - Model parameters

Returns:

Gridable parameters pulled out of the parms

startGridSearch

public Grid.GridSearch startGridSearch(MP params,
                              java.util.Map<java.lang.String,java.lang.Object[]> hyperSearch)

Parameters:

params - Default parameters for grid search builder

hyperSearch - A set of arrays of hyper parameter values, used to specify a simple fully-filled-in grid search.

Returns:

GridSearch Job, with models run with these parameters, built as needed - expected to be an expensive operation. If the models in question are "in progress", a 2nd build will NOT be kicked off. This is a non-blocking call.

remove_impl

protected Futures remove_impl(Futures fs)

Description copied from class: Keyed

Override to remove subparts, but not self, of composite Keyed objects. Examples include Vec (removing associated Chunk keys) and Frame (removing associated Vec keys.)

Overrides:

remove_impl in class Keyed<G extends Grid<MP,G>>

checksum_impl

protected long checksum_impl()

Description copied from class: Keyed

High-quality 64-bit checksum of the content of the object. Similar to hashcode(), but a long to reduce the chance of hash clashes. For composite objects this should be defined using the subcomponents' checksums (or hashcodes if not available). If two Keyed objects have the same checksum() there should be a 1 - 1/2^64 chance that they are the same object by value.

Specified by:

checksum_impl in class Keyed<G extends Grid<MP,G>>