Hello,
the former training API just supported the iterations and cutoff
parameter, but we extended it to support algorithm defined parameters.
These are basically properties which are passed down to the actual trainer.
In our APIs these properties are encapsulated by a TrainingParameters
object.
We deprecated all the old methods which take the iteration and cutoff as
parameter.
Jörn
On 07/03/2012 12:43 PM, Daniel wrote:
Thank for your answer Svetoslav, but I am wondering if I could use simply:
AdaptiveFeatureGenerator adaptFG = null;
model = NameFinderME.train("es", "organization", sampleStream,
ModelUtil.createTrainingParameters(500, 5),
adaptFG ,Collections.<String,
Object>emptyMap());
or I need define the AdaptiveFeatureGenerator as you said before?
2012/7/3, Svetoslav Marinov <[email protected]>:
You can check the API for the current method, otherwise you can use the
following:
NameFinderME.train("es", "organization", sampleStream, featureGenerator,
Collections.<String, Object>emptyMap(), 500, 5)
Where a featureGenerator can for example be defined as:
AdaptiveFeatureGenerator featureGenerator = new
CachedFeatureGenerator(
new AdaptiveFeatureGenerator[]{
new WindowFeatureGenerator(new
TokenFeatureGenerator(), 4, 2),
new WindowFeatureGenerator(new
TokenClassFeatureGenerator(true), 4, 2),
new TokenClassFeatureGenerator(),
new TokenPatternFeatureGenerator(),
new OutcomePriorFeatureGenerator(),
new PreviousMapFeatureGenerator(),
new BigramNameFeatureGenerator(),
new SentenceFeatureGenerator(true, true)
});
Svetoslav
On 2012-07-03 11:53, "Daniel" <[email protected]> wrote:
Hi guys,
Im using NameFinderME.train("es", "organization", sampleStream,
Collections.<String, Object>emptyMap(), 500, 5) to train my model (as
it says in the Doc), but the method is deprecated, any idea about what
method should I use?
Thanks.
