hex.grid

Class HyperSpaceSearchCriteria.RandomDiscreteValueSearchCriteria

java.lang.Object

water.Iced

hex.grid.HyperSpaceSearchCriteria

hex.grid.HyperSpaceSearchCriteria.RandomDiscreteValueSearchCriteria

All Implemented Interfaces:

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

Enclosing class:

HyperSpaceSearchCriteria

public static final class HyperSpaceSearchCriteria.RandomDiscreteValueSearchCriteria
extends HyperSpaceSearchCriteria

Search criteria for a hyperparameter search including directives for how to search and when to stop the search.

NOTE: client ought to call set_default_stopping_tolerance_for_frame(Frame) to get a reasonable stopping tolerance, especially for small N.

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from class hex.grid.HyperSpaceSearchCriteria

HyperSpaceSearchCriteria.CartesianSearchCriteria, HyperSpaceSearchCriteria.RandomDiscreteValueSearchCriteria, HyperSpaceSearchCriteria.SequentialSearchCriteria, HyperSpaceSearchCriteria.StoppingCriteria, HyperSpaceSearchCriteria.Strategy

Field Summary

Fields inherited from class hex.grid.HyperSpaceSearchCriteria

_strategy

Constructor Summary

Constructors

Constructor and Description
RandomDiscreteValueSearchCriteria()

Method Summary

Modifier and Type	Method and Description
static double	default_stopping_tolerance_for_frame(Frame frame) Calculate a reasonable stopping tolerance for the Frame.
int	max_models() Max number of models to build.
double	max_runtime_secs() Max runtime for the entire grid, in seconds.
long	seed() Seed for the random choices of hyperparameter values.
void	set_default_stopping_tolerance_for_frame(Frame frame)
void	set_max_models(int max_models)
void	set_max_runtime_secs(double max_runtime_secs)
void	set_seed(long seed)
void	set_stopping_metric(ScoreKeeper.StoppingMetric stopping_metric)
void	set_stopping_rounds(int stopping_rounds)
void	set_stopping_tolerance(double stopping_tolerance)
ScoreKeeper.StoppingMetric	stopping_metric() Metric to use for convergence checking; only for _stopping_rounds > 0
int	stopping_rounds() Early stopping based on convergence of stopping_metric.
double	stopping_tolerance() Relative tolerance for metric-based stopping criterion: stop if relative improvement is not at least this much.
HyperSpaceSearchCriteria.StoppingCriteria	stoppingCriteria()

Methods inherited from class hex.grid.HyperSpaceSearchCriteria

make, strategy

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

Constructor Detail

RandomDiscreteValueSearchCriteria

public RandomDiscreteValueSearchCriteria()

Method Detail

stoppingCriteria

public HyperSpaceSearchCriteria.StoppingCriteria stoppingCriteria()

Overrides:

stoppingCriteria in class HyperSpaceSearchCriteria

seed

public long seed()

Seed for the random choices of hyperparameter values. Set to a value other than -1 to get a repeatable pseudorandom sequence.

max_models

public int max_models()

Max number of models to build.

max_runtime_secs

public double max_runtime_secs()

Max runtime for the entire grid, in seconds. Set to 0 to disable. Can be combined with max_runtime_secs in the model parameters. If max_runtime_secs is not set in the model parameters then each model build is launched with a limit equal to the remainder of the grid time. If max_runtime_secs is set in the mode parameters each build is launched with a limit equal to the minimum of the model time limit and the remaining time for the grid.

stopping_rounds

public int stopping_rounds()

Early stopping based on convergence of stopping_metric. Stop if simple moving average of the stopping_metric does not improve by stopping_tolerance for k scoring events. Can only trigger after at least 2k scoring events. Use 0 to disable.

stopping_metric

public ScoreKeeper.StoppingMetric stopping_metric()

Metric to use for convergence checking; only for _stopping_rounds > 0

stopping_tolerance

public double stopping_tolerance()

Relative tolerance for metric-based stopping criterion: stop if relative improvement is not at least this much.

default_stopping_tolerance_for_frame

public static double default_stopping_tolerance_for_frame(Frame frame)

Calculate a reasonable stopping tolerance for the Frame. Currently uses only the NA percentage and nrows, but later can take into account the response distribution, response variance, etc.

1/Math.sqrt((1 - frame.naFraction()) * frame.numRows())

set_default_stopping_tolerance_for_frame

public void set_default_stopping_tolerance_for_frame(Frame frame)

set_seed

public void set_seed(long seed)

set_max_models

public void set_max_models(int max_models)

set_max_runtime_secs

public void set_max_runtime_secs(double max_runtime_secs)

set_stopping_rounds

public void set_stopping_rounds(int stopping_rounds)

set_stopping_metric

public void set_stopping_metric(ScoreKeeper.StoppingMetric stopping_metric)

set_stopping_tolerance

public void set_stopping_tolerance(double stopping_tolerance)