Package calhoun.analysis.crf

the interface, main Conrad class, and solver for the Conrad engine.

See:
Description

Interface Summary

CRFInference	an interface to inference algorithms for CRFs.
CRFObjectiveFunctionGradient	an interface to algorithms that compute an objective function and its gradient for CRFs.
CRFTraining	an interface to numerical solvers for optimizing the CRF objective function.
FeatureList	used to pass feature evaluations from a FeatureManager to the Conrad engine during an evaluate call.
FeatureManager<InputType>	evaluates CRF feature functions on the model.
FeatureManagerEdge<InputType>	holds features whose value depends on the current state and the previous state.
FeatureManagerEdgeExplicitLength<InputType>	Since Features are computed as needed dynamically at runtime, a FeatureManager controls training and dynamic creation of the features.
FeatureManagerNode<InputType>	holds features whose value only depends on the current state, and not the previous state.
FeatureManagerNodeBoundaries<InputType>	An extension of FeatureManagerNode for situations where the nodes use in semi-markov models is non-standard.
FeatureManagerNodeExplicitLength<InputType>	Since Features are computed as needed dynamically at runtime, a FeatureManager controls training and dynamic creation of the features.
LocalPathSimilarityScore	interface for local scoring functions used by the local score gradient functions.
ModelManager	overall itnerface for an entire model.

Class Summary

AbstractFeatureManager<InputType>	base class for feature implementations.
BeanModel	a useful base class for creating model beans.
BeanModel.Edge	an edge in the hidden state diagram.
BeanModel.Node	a hidden state.
CacheStrategySpec
CacheStrategySpec.DenseBoundaryCachingDetails
CacheStrategySpec.DenseBoundaryEntry
CacheStrategySpec.DenseCachingDetails	Used in cases where the feature will return a value at every edge and/or node.
CompositeFeatureManager	a feature manager that combines feature types together.
Conrad	the central class for the Conrad engine.
ConstrainedFeatureManager	a feature manager that combines it's composite feature types together into a single feature.
CRFInference.InferenceResult	holder which contains the results of an inference run.
SemiMarkovSetup	holds additional configuration information used in semi-Markov CRFs.

Enum Summary

CacheStrategySpec.CacheStrategy

Package calhoun.analysis.crf Description

the interface, main Conrad class, and solver for the Conrad engine. The Conrad class is the main entry point for the engine, either as an executable through it's main method or programmatically through it's public interface. Very little goes on in the Conrad class, all real functionality is delegated out to the various beans which are configured in the model. This includes the model itself, I/O handling, and even the training and inference algorithms. This component approach allows Conrad to be very easily configured for different problems and approaches.

The main things to understand in this class as the various FeatureManager interfaces if you plan to implement your own features, and the various CRF... interfaces if you plan to reconfigure the engine or solver.