You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@opennlp.apache.org by Jörn Kottmann <ko...@gmail.com> on 2011/06/06 23:37:11 UTC
Perceptron sequence training for name finder
Hi Jason,
actually I experimented with perceptron sequence training for the name
finder. It turned out that it trains a model which only predicts the outcome
"other" in all cases, expect on the first iteration.
Do you believe perceptron sequence training could work for the name finder?
Thanks,
Jörn
Re: Perceptron sequence training for name finder
Posted by Jason Baldridge <ja...@gmail.com>.
I unfortunately cannot look at this right now as I need to focus on other
things for the time being. But, perhaps you could try to update that code
based on PerceptronTrainer? I'm happy to answer questions.
On Wed, Jun 8, 2011 at 12:50 PM, Jörn Kottmann <ko...@gmail.com> wrote:
> On 6/8/11 7:08 PM, Jason Baldridge wrote:
>
>> Sorry, hadn't looked at that code.
>>
>> What are you using for training the perceptron taggers and NER?
>>
>> We have two classes which can be used to train a perceptron model,
> the one you worked on PerceptronTrainer and one for sequence
> training SimplePerceptronSequenceTrainer.
>
> In the OPENNLP-199 issue I used the PerceptronTrainer
> and in this thread I was speaking about the
> SimplePerceptronSequenceTrainer.
> Which fails to train a name finder model which can predict anything else
> than "other".
>
> Jörn
>
--
Jason Baldridge
Assistant Professor, Department of Linguistics
The University of Texas at Austin
http://www.jasonbaldridge.com
http://twitter.com/jasonbaldridge
Re: Perceptron sequence training for name finder
Posted by Jörn Kottmann <ko...@gmail.com>.
On 6/8/11 7:08 PM, Jason Baldridge wrote:
> Sorry, hadn't looked at that code.
>
> What are you using for training the perceptron taggers and NER?
>
We have two classes which can be used to train a perceptron model,
the one you worked on PerceptronTrainer and one for sequence
training SimplePerceptronSequenceTrainer.
In the OPENNLP-199 issue I used the PerceptronTrainer
and in this thread I was speaking about the SimplePerceptronSequenceTrainer.
Which fails to train a name finder model which can predict anything else
than "other".
Jörn
Re: Perceptron sequence training for name finder
Posted by Jason Baldridge <ja...@gmail.com>.
Sorry, hadn't looked at that code.
What are you using for training the perceptron taggers and NER?
On Wed, Jun 8, 2011 at 3:31 AM, Jörn Kottmann <ko...@gmail.com> wrote:
> I was speaking here about the perceptron *sequence* training code, you
> updated the perceptron training code.
>
> Jörn
>
>
>
> On 6/8/11 6:45 AM, Jason Baldridge wrote:
>
>> Please try with the updated perceptron and let me know how it goes. -Jason
>>
>> On Mon, Jun 6, 2011 at 4:37 PM, Jörn Kottmann<ko...@gmail.com> wrote:
>>
>> Hi Jason,
>>>
>>> actually I experimented with perceptron sequence training for the name
>>> finder. It turned out that it trains a model which only predicts the
>>> outcome
>>> "other" in all cases, expect on the first iteration.
>>>
>>> Do you believe perceptron sequence training could work for the name
>>> finder?
>>>
>>> Thanks,
>>> Jörn
>>>
>>>
>>
>>
>
--
Jason Baldridge
Assistant Professor, Department of Linguistics
The University of Texas at Austin
http://www.jasonbaldridge.com
http://twitter.com/jasonbaldridge
Re: Perceptron sequence training for name finder
Posted by Jörn Kottmann <ko...@gmail.com>.
I was speaking here about the perceptron *sequence* training code, you
updated the perceptron training code.
Jörn
On 6/8/11 6:45 AM, Jason Baldridge wrote:
> Please try with the updated perceptron and let me know how it goes. -Jason
>
> On Mon, Jun 6, 2011 at 4:37 PM, Jörn Kottmann<ko...@gmail.com> wrote:
>
>> Hi Jason,
>>
>> actually I experimented with perceptron sequence training for the name
>> finder. It turned out that it trains a model which only predicts the
>> outcome
>> "other" in all cases, expect on the first iteration.
>>
>> Do you believe perceptron sequence training could work for the name finder?
>>
>> Thanks,
>> Jörn
>>
>
>
Re: Perceptron sequence training for name finder
Posted by Jason Baldridge <ja...@gmail.com>.
Please try with the updated perceptron and let me know how it goes. -Jason
On Mon, Jun 6, 2011 at 4:37 PM, Jörn Kottmann <ko...@gmail.com> wrote:
> Hi Jason,
>
> actually I experimented with perceptron sequence training for the name
> finder. It turned out that it trains a model which only predicts the
> outcome
> "other" in all cases, expect on the first iteration.
>
> Do you believe perceptron sequence training could work for the name finder?
>
> Thanks,
> Jörn
>
--
Jason Baldridge
Assistant Professor, Department of Linguistics
The University of Texas at Austin
http://www.jasonbaldridge.com
http://twitter.com/jasonbaldridge