Implement mechanisms to create customized learning system. More...

Inheritance diagram for python.learner.LearnerBuilder.LearnerBuilder:

Classes
class	CGSolverDescriptor
	CGSolver Descriptor helps to implement fluid interface patter on python it encapsulates functionality concerning creation of the CG-Solver. More...

class	FoldingDescriptor
	Folding Descriptor helps to implement fluid interface patter on python it encapsulates functionality concerning the usage for N-fold cross-validation. More...

class	GridDescriptor
	Grid Descriptor helps to implement fluid interface patter on python it encapsulates functionality concerning creation of the grid. More...

class	SpecificationDescriptor
	TrainingSpecification Descriptor helps to implement fluid interface patter on python it encapsulates functionality concerning creation of the training specification. More...

class	StopPolicyDescriptor
	TrainingStopPolicy Descriptor helps to implement fluid interface patter on python it encapsulates functionality concerning creation of the training stop policy. More...

Public Member Functions
	__init__ (self)
	Default constuctor.

	andGetResult (self)
	Returns the builded learner (regressor or classifier), should be called at the end of construction.

	buildClassifier (self)
	Start building Classifier.

	buildRegressor (self)
	Start building Regressor.

	getCheckpointController (self)
	Returns the checkpoint controller.

	getLearner (self)
	Returns the object of learner subclass, that is currently beeing constructed.

	withCGSolver (self)
	Start description of parameters of CG-Solver for learner.

	withCheckpointController (self, controller)
	Attaches checkpoint controller to the learner.

	withFilesFoldingPolicy (self)
	Signals to use N-fold cross validation from a set of files.

	withGrid (self)
	Start description of parameters of CG-Solver for learner.

	withInitialAlphaFromARFFFile (self, filename)
	Signals to use initial data for alpha vector from ARFF file.

	withProgressPresenter (self, presentor)
	Attaches progress presentor to the learner.

	withRandomFoldingPolicy (self)
	Signals to use N-fold cross validation with random folding rule.

	withSequentialFoldingPolicy (self)
	Signals to use N-fold cross validation with sequential folding rule.

	withSpecification (self)
	Start description of specification parameters for learner.

	withStartingIterationNumber (self, iteration)
	Set the starting iteration number ane return the builder object.

	withStopPolicy (self)
	Start description of parameters of stop-policy for learner.

	withStratifiedFoldingPolicy (self)
	Signals to use N-fold cross validation with stratified folding rule.

	withTestingDataFromARFFFile (self, filename)
	Signals to use data from ARFF file for testing dataset.

	withTestingDataFromCSVFile (self, filename)
	Signals to use data from CSV file for testing dataset.

	withTestingDataFromNumPyArray (self, points, values, name="test")

	withTrainingDataFromARFFFile (self, filename, name="train")
	Signals to use data from ARFF file for training dataset.

	withTrainingDataFromCSVFile (self, filename, name="train")
	Signals to use data from CSV file for training dataset.

	withTrainingDataFromNumPyArray (self, points, values, name="train")

Detailed Description

Implement mechanisms to create customized learning system.

Usage examples

To create a learning system first define if it should be for classification

import from pysgpp.extensions.datadriven.learner.LearnerBuilder as LearnerBuilder
builder = LearnerBuilder()
builder = builder.buildClassifier()

or regression

builder = builder.buildRegressor()

LearnerBuilder is implementing Fluent Interface design pattern it means it operates as an automata, switching in some state where you can set all parameters associated with some category. For example to define the grid parameters you switch the builder into GridDescriptor set with

builder = builder.withGrid()...

and then defines corresponding parameters:

builder = builder.withGrid().withLevel(5).withBorder(Types.BorderTypes.TRAPEZOIDBOUNDARY)

Builder can automatically switches to the next state

builder.withGrid()...withCGSolver().withAccuracy(0.00000001)...

After all parameters are set you can return the constructed learning system with

builder.andGetResult()

The complete construction could look like following:

classifier = builder.buildClassifier()\
                     .withTrainingDataFromARFFFile("./datasets/classifier.train.arff")\
                     .withTestingDataFromARFFFile("./datasets/classifier.test.arff")\
                     .withGrid().withLevel(2)\
                     .withSpecification().withLambda(0.00001).withAdaptPoints(2)\
                     .withStopPolicy().withAdaptiveIterationLimit(1)\
                     .withCGSolver().withImax(500)\
                     .withProgressPresenter(InfoToFile("./presentor.test"))\
                     .andGetResult()

Parameters and where I can set them?

level: Gridlevel - withGrid().withLevel(2)
dim: Griddimension - Dimension is identified from the data set (by calling .withTrainingDataFromARFFFile(...) )
adaptive: Using an adaptive Grid with NUM of refines - .withStopPolicy().withAdaptiveIterationLimit(10)
adapt_points: Number of points in one refinement iteration - .withSpecification().withAdaptPoints(100)
adapt_rate: Percentage of points from all refinable points in one refinement iteration - .withSpecification().withAdaptRate(0.05)
adapt_start: The index of adapt step to begin with - Is know handled by loading the learner with specified iteration level from CheckpointController using checkpointController.loadAll(10)
adapt_threshold: refinement threshold - .withSpecification().withAdaptThreshold(0.003)
mode: Specifies the action to do - Call corresponding method, i.e. applyData, learnData, learnDataWithTest, learnDataWithFolding
zeh: Specifies the action to do - .withSpecification().withIdentityOperator() or .withSpecification().withLaplaceOperator()
foldlevel: specifies the number of sets generated - Is set in FoldingDescriptor: builder.withSequentialFoldingPolicy().withLevel(level)
onlyfoldnum: Run only fold I in n-fold cross-validation. Default: run all - checkpointController.generateFoldValidationJob() generates a set of independent learners and a job script to run as SGE job array. In this way all or individual jobs can be ran either with SGE jobs or in console.
lambda: Lambda - .withSpecification().withLambda(0.00001)
imax: Max number of iterations - .withCGSolver().withImax(500)
accuracy: Specifies the accuracy of the CG-Iteration - .withCGSolver().withAccuracy(0.0001)
max_accuracy: If the norm of the residuum falls below ACCURACY, stop the CG iterations - .withCGSolver().withThreshold(0.0000000001)
data: Filename for the Datafile. - .withTestingDataFromARFFFile("./datasets/classifier.test.arff")
test: File containing the testdata - .withTestingDataFromARFFFile("./datasets/classifier.test.arff")
alpha: Filename for a file containing an alpha-Vector - learner = builder.andGetResult() learner.knowledge = LearnedKnowledgeFileAdapter().load("./alphas.arff")
outfile: Filename where the calculated alphas are stored - .withProgressPresenter(InfoToFile("./presentor.test"))
gnuplot: In 2D case, the generated can be stored in a gnuplot readable format - Some Graphs can now be plotted with InfoToGraph
resolution: Specifies the resolution of the gnuplotfile - Not used, as gnuplot is not yet implemented
stats: In this file the statistics from the test are stored - Can be implemented as subclass from LearnerEventController
polynom: Sets the maximum degree for basis functions - .withGrid().withPolynomialBase(2)
border Enables special border base functions - .withGrid().withBorder(Types.BorderTypes.TRAPEZOIDBOUNDARY)
trapezoid-boundary Enables boundary functions that have a point on the boundary for every inner point (Trapezoid) - .withGrid().withBorder(Types.BorderTypes.TRAPEZOIDBOUNDARY)
complete-boundary Enables boundary functions that have more points on the boundary than inner points - .withGrid().withBorder(Types.BorderTypes.COMPLETEBOUNDARY)
verbose: Provides extra output - Set the suitable LearnerEventController implementation, i.e. .withProgressPresenter(InfoToScreen())
normfile: For all modes that read data via stdin. Normalizes data according to boundaries in FILE - Can be implemented as subclass of ARFFAdapter
reuse: Reuse alpha-values for CG - .withCGSolver().withAlphaReusing()
seed: Random seed used for initializing Is set in FoldingDescriptor: builder.withRandomFoldingPolicy().withSeed(level)
regression: Use regression approach - builder.buildRegressor()
checkpoint: Filename for checkpointing - controller = CheckpointController("classification_job") builder.withCheckpointController(controller)
grid: Filename for Grid-resume - .withGrid.fromFile("gridfile.gz")
epochs_limit: Number of refinement iterations (epochs), MSE of test data have to increase, before refinement will stop - .withStopPolicy().withEpochsLimit(20)
mse_limit If MSE of test data fall below this limit, refinement will stop - .withStopPolicy().withMSELimit(0.0003), also ..withStopPolicy().withAccuracyLimit(0.95) for classification accuracy
grid_limit If the number of points on grid exceed grid_limit, refinement will stop - .withStopPolicy().withGridSizeLimit(40000)

Constructor & Destructor Documentation

◆ init()

python.learner.LearnerBuilder.LearnerBuilder.__init__ ( self )

Default constuctor.

References python.learner.LearnerBuilder.LearnerBuilder.__gridDescriptor, python.uq.sampler.asgc.ASGCSamplerBuilder.ASGCSamplerBuilder.__gridDescriptor, python.learner.LearnerBuilder.LearnerBuilder.__learner, python.learner.LearnerBuilder.LearnerBuilder.__specificationDescriptor, python.learner.LearnerBuilder.LearnerBuilder.__stopPolicyDescriptor, python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.__stopPolicyDescriptor, and python.uq.sampler.asgc.ASGCSamplerBuilder.ASGCSamplerBuilder.__stopPolicyDescriptor.

Member Function Documentation

◆ andGetResult()

python.learner.LearnerBuilder.LearnerBuilder.andGetResult ( self )

Returns the builded learner (regressor or classifier), should be called at the end of construction.

Returns: : Learner (Classifier of Regressor)

References python.learner.LearnerBuilder.LearnerBuilder.__gridDescriptor, python.uq.sampler.asgc.ASGCSamplerBuilder.ASGCSamplerBuilder.__gridDescriptor, python.learner.LearnerBuilder.LearnerBuilder.__learner, and python.learner.LearnerBuilder.LearnerBuilder.__specificationDescriptor.

◆ buildClassifier()

python.learner.LearnerBuilder.LearnerBuilder.buildClassifier ( self )

Start building Classifier.

Returns: : LearnerBuilder itself

References python.learner.LearnerBuilder.LearnerBuilder.__buildCommonLearner(), and python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ buildRegressor()

python.learner.LearnerBuilder.LearnerBuilder.buildRegressor ( self )

Start building Regressor.

Returns: : LearnerBuilder itself

References python.learner.LearnerBuilder.LearnerBuilder.__buildCommonLearner(), and python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ getCheckpointController()

python.learner.LearnerBuilder.LearnerBuilder.getCheckpointController ( self )

Returns the checkpoint controller.

Returns: the checkpoint controller

References python.learner.LearnerBuilder.LearnerBuilder.__checkpointController.

◆ getLearner()

python.learner.LearnerBuilder.LearnerBuilder.getLearner ( self )

Returns the object of learner subclass, that is currently beeing constructed.

Returns: the object of learner subclass, that is currently beeing constructed

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

Referenced by python.learner.LearnerBuilder.LearnerBuilder.GridDescriptor.__getattr__(), python.learner.LearnerBuilder.LearnerBuilder.StopPolicyDescriptor.__getattr__(), python.learner.LearnerBuilder.LearnerBuilder.SpecificationDescriptor.__getattr__(), python.learner.LearnerBuilder.LearnerBuilder.CGSolverDescriptor.__getattr__(), python.learner.LearnerBuilder.LearnerBuilder.FoldingDescriptor.__getattr__(), and python.learner.LearnerBuilder.LearnerBuilder.GridDescriptor.__init__().

◆ withCGSolver()

python.learner.LearnerBuilder.LearnerBuilder.withCGSolver ( self )

Start description of parameters of CG-Solver for learner.

Returns: : CGSolverDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__solverDescriptor, and python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.__solverDescriptor.

◆ withCheckpointController()

python.learner.LearnerBuilder.LearnerBuilder.withCheckpointController	(	self,
		controller
	)

Attaches checkpoint controller to the learner.

Parameters

controller Checkpoint controller which implements LearnerEventController

Returns: : LearnerBuilder

References python.learner.LearnerBuilder.LearnerBuilder.__checkpointController, and python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withFilesFoldingPolicy()

python.learner.LearnerBuilder.LearnerBuilder.withFilesFoldingPolicy ( self )

Signals to use N-fold cross validation from a set of files.

Returns: : FoldingDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__foldingPolicyDescriptor, and python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.__foldingPolicyDescriptor.

◆ withGrid()

python.learner.LearnerBuilder.LearnerBuilder.withGrid ( self )

Start description of parameters of CG-Solver for learner.

Returns: : GridDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__gridDescriptor, and python.uq.sampler.asgc.ASGCSamplerBuilder.ASGCSamplerBuilder.__gridDescriptor.

◆ withInitialAlphaFromARFFFile()

python.learner.LearnerBuilder.LearnerBuilder.withInitialAlphaFromARFFFile	(	self,
		filename
	)

Signals to use initial data for alpha vector from ARFF file.

Parameters

filename Filename where to read the data from

Returns: : LearnerBuilder object itself

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withProgressPresenter()

python.learner.LearnerBuilder.LearnerBuilder.withProgressPresenter	(	self,
		presentor
	)

Attaches progress presentor to the learner.

Parameters

presentor progress presentor which implements LearnerEventController

Returns: : LearnerBuilder

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withRandomFoldingPolicy()

python.learner.LearnerBuilder.LearnerBuilder.withRandomFoldingPolicy ( self )

Signals to use N-fold cross validation with random folding rule.

Returns: : FoldingDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__foldingPolicyDescriptor, and python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.__foldingPolicyDescriptor.

◆ withSequentialFoldingPolicy()

python.learner.LearnerBuilder.LearnerBuilder.withSequentialFoldingPolicy ( self )

Signals to use N-fold cross validation with sequential folding rule.

Returns: : FoldingDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__foldingPolicyDescriptor, and python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.__foldingPolicyDescriptor.

◆ withSpecification()

python.learner.LearnerBuilder.LearnerBuilder.withSpecification ( self )

Start description of specification parameters for learner.

Returns: : SpecificationDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__specificationDescriptor.

◆ withStartingIterationNumber()

python.learner.LearnerBuilder.LearnerBuilder.withStartingIterationNumber	(	self,
		iteration
	)

Set the starting iteration number ane return the builder object.

Parameters

iteration integer starting iteration number

Returns: : LeanreBuilder

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withStopPolicy()

python.learner.LearnerBuilder.LearnerBuilder.withStopPolicy ( self )

Start description of parameters of stop-policy for learner.

Returns: : StopPolicyDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__stopPolicyDescriptor, python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.__stopPolicyDescriptor, and python.uq.sampler.asgc.ASGCSamplerBuilder.ASGCSamplerBuilder.__stopPolicyDescriptor.

Referenced by python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.create().

◆ withStratifiedFoldingPolicy()

python.learner.LearnerBuilder.LearnerBuilder.withStratifiedFoldingPolicy ( self )

Signals to use N-fold cross validation with stratified folding rule.

Returns: : FoldingDescriptor

References python.learner.LearnerBuilder.LearnerBuilder.__foldingPolicyDescriptor, and python.uq.learner.builder.RegressorSpecificationDescriptor.RegressorSpecificationDescriptor.__foldingPolicyDescriptor.

◆ withTestingDataFromARFFFile()

python.learner.LearnerBuilder.LearnerBuilder.withTestingDataFromARFFFile	(	self,
		filename
	)

Signals to use data from ARFF file for testing dataset.

Parameters

filename Filename where to read the data from

Returns: : LearnerBuilder object itself

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withTestingDataFromCSVFile()

python.learner.LearnerBuilder.LearnerBuilder.withTestingDataFromCSVFile	(	self,
		filename
	)

Signals to use data from CSV file for testing dataset.

Parameters

filename Filename where to read the data from

Returns: : LearnerBuilder object itself

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withTestingDataFromNumPyArray()

python.learner.LearnerBuilder.LearnerBuilder.withTestingDataFromNumPyArray	(	self,
		points,
		values,
		name = `"test"`
	)

References python.learner.LearnerBuilder.LearnerBuilder.withTrainingDataFromNumPyArray().

◆ withTrainingDataFromARFFFile()

python.learner.LearnerBuilder.LearnerBuilder.withTrainingDataFromARFFFile	(	self,
		filename,
		name = `"train"`
	)

Signals to use data from ARFF file for training dataset.

Parameters

filename	Filename where to read the data from
name	Category name, default: "train"

Returns: : LearnerBuilder

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withTrainingDataFromCSVFile()

python.learner.LearnerBuilder.LearnerBuilder.withTrainingDataFromCSVFile	(	self,
		filename,
		name = `"train"`
	)

Signals to use data from CSV file for training dataset.

Parameters

filename	Filename where to read the data from
name	Category name, default: "train"

Returns: : LearnerBuilder

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

◆ withTrainingDataFromNumPyArray()

python.learner.LearnerBuilder.LearnerBuilder.withTrainingDataFromNumPyArray	(	self,
		points,
		values,
		name = `"train"`
	)

References python.learner.LearnerBuilder.LearnerBuilder.__learner.

Referenced by python.learner.LearnerBuilder.LearnerBuilder.withTestingDataFromNumPyArray().

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

datadriven/python/learner/LearnerBuilder.py

Classes

Public Member Functions

Detailed Description

Usage examples

Parameters and where I can set them?

Constructor & Destructor Documentation

◆ __init__()

Member Function Documentation

◆ andGetResult()

◆ buildClassifier()

◆ buildRegressor()

◆ getCheckpointController()

◆ getLearner()

◆ withCGSolver()

◆ withCheckpointController()

◆ withFilesFoldingPolicy()

◆ withGrid()

◆ withInitialAlphaFromARFFFile()

◆ withProgressPresenter()

◆ withRandomFoldingPolicy()

◆ withSequentialFoldingPolicy()

◆ withSpecification()

◆ withStartingIterationNumber()

◆ withStopPolicy()

◆ withStratifiedFoldingPolicy()

◆ withTestingDataFromARFFFile()

◆ withTestingDataFromCSVFile()

◆ withTestingDataFromNumPyArray()

◆ withTrainingDataFromARFFFile()

◆ withTrainingDataFromCSVFile()

◆ withTrainingDataFromNumPyArray()

◆ init()