Thank you! Yeah i meant the first.. But the second sounds like a good workaround. Isola maxent the default algorithm?
Il 30 lug 2017 4:40 PM, "Cohan Sujay Carlos" <[email protected]> ha scritto: > If your question is "Can we set the relative importance of features for > classification using an OpenNLP interface," I believe the answer is that > there isn't. > > But if your question is, "Can we set the relative importance of features > for classification using some means?" then the answer is yes. > > You can do it by repeating the feature the required number of times in your > training file. > > Eg: *CLASS2* token1 token2 token3 token4^2 can be written as *CLASS2* > token1 token2 token3 token4 token4 > > I don't know if this will lead to any performance improvements in the case > of the maxent classification algorithm, but if done right, it could improve > the performance with naive bayes. > > Cohan > > On Sun, Jul 30, 2017 at 7:14 PM, Damiano Porta <[email protected]> > wrote: > > > anyone? > > > > 2017-07-26 18:14 GMT+02:00 Damiano Porta <[email protected]>: > > > > > Hello, > > > I would like to boost specific features/tokens on the training set. > Can i > > > set the weights somehow? > > > Something like: > > > > > > *CLASS1* token1^10 token2 token3^5 > > > *CLASS2* token1 token2 token3 token4^5 > > > > > > Thanks > > > Damiano > > > > > >
