Hello, I updated the code and afterwards spent some time evaluating it again. The maxent training is very close to our maxent classifier. I also checked the training code again and it looks good to me, but it would be nice if you can review it.
There are a couple of other classifiers in mallet, it should be trivial to expose them all to OpenNLP. Jörn On Tue, Oct 20, 2015 at 9:12 AM, Rodrigo Agerri <[email protected]> wrote: > Hello, > > Thanks. I thought I had an idea for CRF not obtaining good results > with OpenNLP default features, e.g., > > http://lingpipe-blog.com/2006/11/22/why-do-you-hate-crfs/ > > but if results are also worse in Maxent, that is intriguing. I will > look at the Mallet implementation to see if I find out something. > > R > > > > On Mon, Oct 12, 2015 at 4:07 PM, Joern Kottmann <[email protected]> > wrote: > > Hello, > > > > fixed up the code a bit. The performance is not really good. Do you have > > any idea why that could be? > > > > Neither the maxent or crf get good evaluation numbers on NER. > > > > I will push the changes and then you can experiment with it too. > > > > Jörn > > > > > > On Mon, Oct 5, 2015 at 4:45 PM, Rodrigo Agerri <[email protected]> > wrote: > > > >> Hi, > >> > >> On Tue, Sep 29, 2015 at 3:41 PM, Joern Kottmann <[email protected]> > >> wrote: > >> > We can also move > >> > it to the sandbox, releasing it at Apache might be more difficult > since > >> > mallet pulls in incompatible licensed dependencies. But maybe that > >> changed, > >> > we can check. > >> > >> Mallet is released under Common Public License > >> > >> http://opensource.org/licenses/cpl1.0.php > >> > >> but as you have mentioned, it pulls several dependencies that are > >> LGPL. These are the dependencies: > >> > >> <dependency> > >> <groupId>org.beanshell</groupId> > >> <artifactId>bsh</artifactId> > >> <version>2.0b4</version> > >> </dependency> > >> > >> This version is LGPL, however, later versions are APL 2.0 > >> > >> https://github.com/beanshell/beanshell > >> > >> <dependency> > >> <groupId>jgrapht</groupId> > >> <artifactId>jgrapht</artifactId> > >> <version>0.6.0</version> > >> </dependency> > >> > >> that version was also LGPL, but it has now been dual-licensed with EPL > 1.0 > >> > >> https://github.com/jgrapht/jgrapht/wiki/Relicensing > >> > >> which could be included also in APL 2.0 projects > >> > >> http://www.apache.org/legal/resolved.html > >> > >> <dependency> > >> <groupId>net.sf.jwordnet</groupId> > >> <artifactId>jwnl</artifactId> > >> <version>1.4_rc3</version> > >> </dependency> > >> > >> BSD license, but this library has already been discussed here. > >> > >> <dependency> > >> <groupId>net.sf.trove4j</groupId> > >> <artifactId>trove4j</artifactId> > >> <version>2.0.2</version> > >> </dependency> > >> > >> LGPL-ed. > >> > >> <dependency> > >> <groupId>com.googlecode.matrix-toolkits-java</groupId> > >> <artifactId>mtj</artifactId> > >> <version>0.9.14</version> > >> </dependency> > >> > >> also LGPL > >> > >> Rodrigo > >> >
