Hello again,
Ok so the cross-validator needs an empty Map in order not to throw a
NPE. The correct method is:
public String cross-validate(String fname, int kFolds){
FileInputStream sampleDataIn = new FileInputStream(fname);
ObjectStream<NameSample> sampleStream = new NameSampleDataStream(
new PlainTextByLineStream(
sampleDataIn.getChannel(), "UTF-8"));
TokenNameFinderCrossValidator judge =
new TokenNameFinderCrossValidator("en", null, new
TrainingParameters(), null, new HashMap(), null);
judge.evaluate(sampleStream, kFolds);
FMeasure result = evaluator.getFMeasure();
return result.toString();
}
However if i try to pass a "NameEvaluationErrorListener" as the last
parameter instead of null the cross-validator just hangs before i can
do getFMeasure()!!! Actually it hangs right after printing the
misclassifications for the last document in the training data...If i
leave the listeners parameter as null or with a different listener (e.g
TokenNameFinderDetailedFMeasureListener) it works just fine meaning i
can get my FMeasure and print it...
Any ideas as to what could be the problem with the
NameEvaluationErrorListener which is indeed the most useful listener for
cross-validation?
Thanks in advance...
Jim
On 09/03/12 17:34, [email protected] wrote:
Hi,
Although the JavaDoc states that you can pass in a null resource, looks
like it causes the null pointer exception later.
Please try passing an empty map instead of null.
Open a Jira issue so we can improve it.
Thank you,
William
On Thu, Mar 8, 2012 at 6:28 PM, Jim - FooBar();<[email protected]>wrote:
I should clarify that i 'm using the latest code from svn...
Any ideas?
On 08/03/12 16:10, Dimitrios wrote:
Hi everyone,
Can someone please explain why i 'm getting this NPE when trying to use
the TokenNameFinderCrossValidator?
TokenNameFinderEvaluator seems to be working just fine but because i've
got very little training-data i'd like to do cross-validation as well...
*The code is as follows:*
public String cross-validate(String fname, int kFolds){
FileInputStream sampleDataIn = new FileInputStream(fname);
ObjectStream<NameSample> sampleStream = new NameSampleDataStream(
new PlainTextByLineStream(
sampleDataIn.getChannel(), "UTF-8"));
TokenNameFinderCrossValidator judge = new TokenNameFinderCrossValidator(*
*"en",
null,
new TrainingParameters(),
null,
null,
null);
judge.evaluate(sampleStream, kFolds);
FMeasure result = evaluator.getFMeasure();
return result.toString();
}
*The exception is thrown right after the first 100 iterations for the
first fold*:
NullPointerException opennlp.tools.namefind.**TokenNameFinderModel.<init>
(TokenNameFinderModel.java:90)
*If i omit the "new NameSampleDataStream()" (as shown in the docs) i get
a different error saying:*
ClassCastException java.lang.String cannot be cast to
opennlp.tools.namefind.**NameSample opennlp.tools.namefind.**
TokenNameFinderCrossValidator$**NameToDocumentSampleStream.**read (**
TokenNameFinderCrossValidator.**java:67)
What am i missing?
Thanks in advance...
Jim
