You are viewing a plain text version of this content. The canonical link for it is here.

Posted to users@opennlp.apache.org by Fraser Bowen <fr...@westernacher.com> on 2018/08/03 14:24:06 UTC

Training from previously trained checkpoints

Dear OpenNLP community,


I would like to know if it is possible to train a Name Finder model using a previously trained model as a starting point? I haven't found a "train_from" feature, and I wonder if having training checkpoints would at all be feasible.


Many thanks in advance!

Fraser

Re: Training from previously trained checkpoints

Posted by Dan Russ <da...@gmail.com>.

Hi Fraser,  
   This questions comes up a lot.  I do not believe it is possible.  The models don’t store the training data, and the NameFinder does not an “online” training method.  You may try to download training data (like CoNLL 2002?) and append your data in the appropriate format. 
Sorry,
Daniel

> On Aug 3, 2018, at 10:24 AM, Fraser Bowen <fr...@westernacher.com> wrote:
> 
> Dear OpenNLP community,
> 
> 
> I would like to know if it is possible to train a Name Finder model using a previously trained model as a starting point? I haven't found a "train_from" feature, and I wonder if having training checkpoints would at all be feasible.
> 
> 
> Many thanks in advance!
> 
> Fraser