Conditional Random Field (CRF)

Prev | Home | Next

6. Configuration Parameters

There are certain parameters that can be accepted by the iitb.CRF package. While creating an instance of the CRF class, you will have to pass these parameters as a java.util.Properties object. You might want to specify certain parameters like numLabels and modelGraph which will be used to create the graphical model for your application as described in the Models section. Some of the parameters which are commonly used are described below. Refer the source code for the advanced parameters.

maxIters: Number of training iterations over the training set; default set to 100
trainer:Training algorithms: default is the standard maximum log-likelihood trainer; you can specify "Collins" for Discriminative training of the model
: A sample dataset and the corresponding configuration file
numLabels:Number of labels
modelGraph: As described in the Models section

Prev | Home | Next

Copyright © 2004 KReSIT, IIT Bombay. All rights reserved