Uses the Efficient Global Optimization (EGO) algorithm to optimize in hyperparameter space. More...

Inheritance diagram for EGOModelBuilder:

Public Member Functions
function	EGOModelBuilder (var config)
	Class constructor. More...

function	runLoop ()
	Uses EGO to generate new models. More...

function	createMovie ()
	Creates a movie of the best models versus the sample or model iteration. More...

function	defaultFitnessFunction (var pop, var train)
	Return the score of the given models or parameter vector (representing one or more models). More...

function	generateNewModels (var number, var wantModels, var previousPop)
	Generates starting points for the hyperparameter optimization. More...

function	observe (var tag, var x_value, var object)
	Observe each observable. More...

function	printBestResults ()
	Print information about the best modeling results so far. More...

function	rebuildBestModel (var keepOldModels)
	Each time new samples are added, the last x best models are re-evaluated against all measures. More...

function	registerObserver (var tag, var x_name, var description, var observables)
	Register a new observable. More...

function	scoreModels (var models)
	Attaches a score to a given set of models. More...

function	setData (var state)
	Set the data point (samples and values) on the ModelBuilder object. More...

function	setModelInfo (var model, var score)
	Set important contextual information on the model: the input names, output names, transformation values, measure data, etc. More...

function	done ()
	This function returns true if all final targets were met. More...

function	getBestModel ()
	Return the last best model built. More...

function	getBestModelScore ()
	Get the Global score of the best model. More...

function	getBestModels (var n)
	Return the `n' last best models, optionally filtered for one output. More...

function	getData ()
	Get current samples en values, i.e. More...

function	getState ()
	Get the complete current state structure. More...

function	getKeepOldModels ()
	Should old models be kept when managing the model trace. More...

function	clearBestModels ()
	Clears the best model trace. More...

function	getMaximumTime ()
	Get the maximum running time. More...

function	getOutputDirectory ()
	Get the output directory. More...

function	getModelFactory ()
	Get the ModelFactory. More...

function	getNumObjectives ()
	Returns the number of objectives the model builder has to deal with (if they are not combined) This corresponds to the number of enabled measures. More...

function	getOutputNames ()
	Returns the output names. More...

function	getParallelMode ()
	Returns whether we are running in parallel mode. More...

function	getParetoMode ()
	Returns whether we are doing multiobjective optimization. More...

function	getRestartStrategy ()
	Returns the restart strategy we are using. More...

function	getStartTime ()
	Returns the time we started. More...

function	isSamplingEnabled ()
	Returns whether adaptive sampling is enabled. More...

function	setLevelPlotObject (var levelObj)
	Sets a LevelPlot object. More...

function	setModelFactory (var mf)
	Sets a ModelFactory object. More...

function	setRunNumber (var number)
	Sets the run number. More...

function	setStartTime (var time, var maxTime)
	Sets the start time. More...

function	evaluateMeasures (var model)
	This function evaluates the model on all the measures and returns the scores. More...

Detailed Description

Uses the Efficient Global Optimization (EGO) algorithm to optimize in hyperparameter space.

Constructor & Destructor Documentation

function EGOModelBuilder ( var config )

inline

Class constructor.

Returns: instance of the class

Member Function Documentation

function clearBestModels ( )

inlineinherited

Clears the best model trace.

function createMovie ( )

inherited

Creates a movie of the best models versus the sample or model iteration.

Create a quicktime movie of all the .jpeg files in the output directory.

function defaultFitnessFunction	(	var	pop,
		var	train
	)

inherited

Return the score of the given models or parameter vector (representing one or more models).

This is a generic implementation that every model builder can use.

Parameters

pop	population of sumo models
train	do we want to fit the sumo models (boolean)

function done ( )

inlineinherited

This function returns true if all final targets were met.

function evaluateMeasures ( var model )

inherited

This function evaluates the model on all the measures and returns the scores.

It also calculates the global score of the model, based on a combination of the (weighted) separate measure scores.

Parameters

model sumo Model

function generateNewModels	(	var	number,
		var	wantModels,
		var	previousPop
	)

inherited

Generates starting points for the hyperparameter optimization.

Also implements different restart strategies (= i.e.

Parameters

number	number of points to generate
wantModels	wrap the points in sumo models (boolean)
previousPop	previous sumo population

how to continue the hyperparameter optimization process in the next modeling iteration)

function getBestModel ( )

inlineinherited

Return the last best model built.

function getBestModels ( var n )

inlineinherited

Return the `n' last best models, optionally filtered for one output.

Parameters

n	number of sumo models to retrieve

function getBestModelScore ( )

inlineinherited

Get the Global score of the best model.

function getData ( )

inlineinherited

Get current samples en values, i.e.

the samples and values of all simulations ran up till now.

function getKeepOldModels ( )

inlineinherited

Should old models be kept when managing the model trace.

function getMaximumTime ( )

inlineinherited

Get the maximum running time.

function getModelFactory ( )

inlineinherited

Get the ModelFactory.

function getNumObjectives ( )

inlineinherited

Returns the number of objectives the model builder has to deal with (if they are not combined) This corresponds to the number of enabled measures.

function getOutputDirectory ( )

inlineinherited

Get the output directory.

function getOutputNames ( )

inlineinherited

Returns the output names.

function getParallelMode ( )

inlineinherited

Returns whether we are running in parallel mode.

function getParetoMode ( )

inlineinherited

Returns whether we are doing multiobjective optimization.

function getRestartStrategy ( )

inlineinherited

Returns the restart strategy we are using.

function getStartTime ( )

inlineinherited

Returns the time we started.

function getState ( )

inlineinherited

Get the complete current state structure.

function isSamplingEnabled ( )

inlineinherited

Returns whether adaptive sampling is enabled.

function observe	(	var	tag,
		var	x_value,
		var	object
	)

inherited

Observe each observable.

Records the values in the matching profilers.

Parameters

tag	profiles to write to (e.g., all profilers starting with 'best')
x_value	x value to profile
object	y value to profile

function printBestResults ( )

inherited

Print information about the best modeling results so far.

Includes the best model score, type of error function, target, etc.

function rebuildBestModel ( var keepOldModels )

inherited

Each time new samples are added, the last x best models are re-evaluated against all measures.

rebuildBestModels ensures lucky models (on the new data) to be kept.

Parameters

keepOldModels keep the old model trace (boolean)

The best of these is set as the new best model

function registerObserver	(	var	tag,
		var	x_name,
		var	description,
		var	observables
	)

inherited

Register a new observable.

Only registers the profiler if it is enabled in the configuration.

Parameters

tag	profiles to write to (e.g., all profilers starting with 'best')
x_name	name of the x-axis
description	description of the observable
observables	cell array of Observable objects

function runLoop ( )

Uses EGO to generate new models.

TODO.

function scoreModels ( var models )

inherited

Attaches a score to a given set of models.

Calls all Measure object attached to the ModelBuilder.

Parameters

models cell array of sumo models

Afterwards, the sumo models are ordered based on the score.

function setData ( var state )

inherited

Set the data point (samples and values) on the ModelBuilder object.

New data for a ModelBuilder passes through here.

Parameters

state struct

function setLevelPlotObject ( var levelObj )

inlineinherited

Sets a LevelPlot object.

Parameters

levelObj LevelPlot object

function setModelFactory ( var mf )

inlineinherited

Sets a ModelFactory object.

Parameters

mf	ModelFactory object

function setModelInfo	(	var	model,
		var	score
	)

inherited

Set important contextual information on the model: the input names, output names, transformation values, measure data, etc.

This is information only the modelbuilder has.

Parameters

model	sumo Model
score	score

function setRunNumber ( var number )

inlineinherited

Sets the run number.

Parameters

number run number.

function setStartTime	(	var	time,
		var	maxTime
	)

inlineinherited

Sets the start time.

Parameters

time	start time
maxTime	maximum time we are allowed to run

The documentation for this class was generated from the following files:

src/matlab/modelbuilders/@EGOModelBuilder/EGOModelBuilder.m
src/matlab/modelbuilders/@EGOModelBuilder/runLoop.m

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation