com.bayesserver.learning.parameters

Class ParameterLearning

java.lang.Object

com.bayesserver.learning.parameters.ParameterLearning

public final class ParameterLearning
extends Object

Learns the parameters of Bayesian networks and Dynamic Bayesian networks, from data. See learn(com.bayesserver.data.EvidenceReaderCommand, com.bayesserver.learning.parameters.ParameterLearningOptions) for sample code.

Constructor Summary

Constructors

Constructor and Description
ParameterLearning(Network network, InferenceFactory factory) Initializes a new instance of the ParameterLearning class.

Method Summary

Modifier and Type	Method and Description
Network	getNetwork() Returns the relevant network.
ParameterLearningOutput	learn(EvidenceReaderCommand readerCommand, List<DistributionSpecification> distributionSpecifications, ParameterLearningOptions options) Learns the parameters of a Bayesian network or Dynamic Bayesian network, from data.
ParameterLearningOutput	learn(EvidenceReaderCommand readerCommand, ParameterLearningOptions options) Learns the parameters of a Bayesian network or Dynamic Bayesian network, from data.
static ParameterLearningOutput	learnDistributed(Network network, List<DistributionSpecification> distributionSpecifications, ParameterLearningOptions options, Distributer<DistributerContext> distributer) Learns the parameters of a Bayesian network or Dynamic Bayesian network from data, on a distributed platform.
static ParameterLearningOutput	learnDistributed(Network network, ParameterLearningOptions options, Distributer<DistributerContext> distributer) Learns the parameters of a Bayesian network or Dynamic Bayesian network from data, on a distributed platform.
static void	learnDistributedMapper(EvidencePartition<DistributedMapperContext> partition, NameValuesReader configuration, NameValuesWriter output, InferenceFactory factory) This method should be called during distributed parameter learning on a distributed partition.
static void	learnDistributedReducer(Iterable<NameValuesReader> inputs, NameValuesReader configuration, NameValuesWriter output) Aggregates (reduces) the results obtained from the distributed calls to learnDistributedMapper(com.bayesserver.data.distributed.EvidencePartition<com.bayesserver.learning.parameters.DistributedMapperContext>, com.bayesserver.NameValuesReader, com.bayesserver.NameValuesWriter, com.bayesserver.inference.InferenceFactory).

Methods inherited from class java.lang.Object

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

Constructor Detail

ParameterLearning

public ParameterLearning(Network network,
                         InferenceFactory factory)

Initializes a new instance of the ParameterLearning class. Learning uses inference as a subroutine, and creates one or more inference engines via the [factory] parameter.

Parameters:

network - The network whose parameters are being learnt.

factory - The inference factory used to create inference engines in cases when learning requires inference.

Method Detail

getNetwork

public Network getNetwork()

Returns the relevant network.

Returns:

The network.

learn

public ParameterLearningOutput learn(EvidenceReaderCommand readerCommand,
                                     ParameterLearningOptions options)

Learns the parameters of a Bayesian network or Dynamic Bayesian network, from data.

Parameters:

readerCommand - Can create a reader containing the data to learn from.

options - Learning options.

Returns:

Summary information about parameter learning.

learnDistributedReducer

public static void learnDistributedReducer(Iterable<NameValuesReader> inputs,
                                           NameValuesReader configuration,
                                           NameValuesWriter output)

Aggregates (reduces) the results obtained from the distributed calls to learnDistributedMapper(com.bayesserver.data.distributed.EvidencePartition<com.bayesserver.learning.parameters.DistributedMapperContext>, com.bayesserver.NameValuesReader, com.bayesserver.NameValuesWriter, com.bayesserver.inference.InferenceFactory).

Parameters:

inputs - The results from each distributed mapper.

configuration - Configuration data that was distributed during the call to learnDistributed(com.bayesserver.Network, com.bayesserver.learning.parameters.ParameterLearningOptions, com.bayesserver.Distributer<com.bayesserver.learning.parameters.DistributerContext>).

output - The aggregated results which will be used by the driver program which called learnDistributed(com.bayesserver.Network, com.bayesserver.learning.parameters.ParameterLearningOptions, com.bayesserver.Distributer<com.bayesserver.learning.parameters.DistributerContext>).

learnDistributedMapper

public static void learnDistributedMapper(EvidencePartition<DistributedMapperContext> partition,
                                          NameValuesReader configuration,
                                          NameValuesWriter output,
                                          InferenceFactory factory)

This method should be called during distributed parameter learning on a distributed partition.

Parameters:

partition - The distributed data to be learnt from.

configuration - Configuration data that was distributed during the call to learnDistributed(com.bayesserver.Network, com.bayesserver.learning.parameters.ParameterLearningOptions, com.bayesserver.Distributer<com.bayesserver.learning.parameters.DistributerContext>).

output - Stores the result of the distributed process. These results will later be passed to an aggregator (reducer).

factory - The inference factory used to create inference engines in cases when learning requires inference.

learnDistributed

public static ParameterLearningOutput learnDistributed(Network network,
                                                       ParameterLearningOptions options,
                                                       Distributer<DistributerContext> distributer)

Learns the parameters of a Bayesian network or Dynamic Bayesian network from data, on a distributed platform.

Parameters:

network - The network whose parameters are being learnt.

options - Learning options.

distributer - The caller must implement this interface, which allows processing to be distributed. learnDistributedMapper(com.bayesserver.data.distributed.EvidencePartition<com.bayesserver.learning.parameters.DistributedMapperContext>, com.bayesserver.NameValuesReader, com.bayesserver.NameValuesWriter, com.bayesserver.inference.InferenceFactory) should be called in the map phase, andlearnDistributedReducer(java.lang.Iterable<com.bayesserver.NameValuesReader>, com.bayesserver.NameValuesReader, com.bayesserver.NameValuesWriter) should be called to reduce the results.

Returns:

Summary information about parameter learning.

learnDistributed

public static ParameterLearningOutput learnDistributed(Network network,
                                                       List<DistributionSpecification> distributionSpecifications,
                                                       ParameterLearningOptions options,
                                                       Distributer<DistributerContext> distributer)

Learns the parameters of a Bayesian network or Dynamic Bayesian network from data, on a distributed platform. See com.bayesserver.learning.parameters.ParameterLearning#learn(IEvidenceReader, ParameterLearningOptions) for sample code.

Parameters:

network - The network whose parameters are being learnt.

distributionSpecifications - The distributions to learn and options.

options - Learning options.

distributer - The caller must implement this interface, which allows processing to be distributed. learnDistributedMapper(com.bayesserver.data.distributed.EvidencePartition<com.bayesserver.learning.parameters.DistributedMapperContext>, com.bayesserver.NameValuesReader, com.bayesserver.NameValuesWriter, com.bayesserver.inference.InferenceFactory) should be called in the map phase, andlearnDistributedReducer(java.lang.Iterable<com.bayesserver.NameValuesReader>, com.bayesserver.NameValuesReader, com.bayesserver.NameValuesWriter) should be called to reduce the results.

Returns:

Summary information about parameter learning.

learn

public ParameterLearningOutput learn(EvidenceReaderCommand readerCommand,
                                     List<DistributionSpecification> distributionSpecifications,
                                     ParameterLearningOptions options)

Learns the parameters of a Bayesian network or Dynamic Bayesian network, from data. The parameters of the Network are replaced during learning. If necessary, a copy of the original network can be made prior to learning, to keep the original values.

Parameters:

readerCommand - Can create a reader containing the data to learn from.

distributionSpecifications - The distributions to learn and options.

options - Learning options.

Returns:

Summary information about parameter learning.