opennlp.maxent

Interface Evalable

public interface Evalable

Interface for components which use maximum entropy models and can evaluate the performace of the models using the TrainEval class.

Version:

$Revision: 1.2 $, $Date: 2010/09/06 08:02:18 $

Author:

Gann Bierner

Method Summary

Modifier and Type	Method and Description
opennlp.model.EventCollector	getEventCollector(java.io.Reader r) Returns the EventCollector that is used to collect all relevant information from the data file.
java.lang.String	getNegativeOutcome() The outcome that should be considered a negative result.
void	localEval(opennlp.model.MaxentModel model, java.io.Reader r, Evalable e, boolean verbose) If the -l option is selected for evaluation, this method will be called rather than TrainEval's evaluation method.

Method Detail

getNegativeOutcome

java.lang.String getNegativeOutcome()

The outcome that should be considered a negative result. This is used for computing recall. In the case of binary decisions, this would be the false one.

Returns:

the events that this EventCollector has gathered

getEventCollector

opennlp.model.EventCollector getEventCollector(java.io.Reader r)

Returns the EventCollector that is used to collect all relevant information from the data file. This is used for to test the predictions of the model. Note that if some of your features are the oucomes of previous events, this method will give you results assuming 100% performance on the previous events. If you don't like this, use the localEval method.

Parameters:

r - A reader containing the data for the event collector

Returns:

an EventCollector

localEval

void localEval(opennlp.model.MaxentModel model,
               java.io.Reader r,
               Evalable e,
               boolean verbose)

If the -l option is selected for evaluation, this method will be called rather than TrainEval's evaluation method. This is good if your features includes the outcomes of previous events.

Parameters:

model - the maxent model to evaluate

r - Reader containing the data to process

e - The original Evalable. Probably not relevant.

verbose - a request to print more specific processing information