--Hi

I know that there are some discussion about the use of MaxEnt with
real-valued features out there but this question is about its performance.

So, I used MaxEnt with binary features in some problems (multiclass
classification problems such as NER) and it worked quite well in terms of
performance (accuracy, F-measures...). I recently want to add some real
valued features to the systems. So, my feature set will include both binary
and real-valued features.

In principle, we can do it in MaxEnt package by assigning binary features
with the values of 1 and keeping the real-valued features unchanged.
However, I wonder if this would be helpful for the final performance. Have
anybody tried this before? Any suggestion would be appreciated.

Thank you very much.

Reply via email to