On 10/08/2014 12:35 PM, Gustavo Knuppe wrote:
Jörn,
IKVM works very well, but uses a lot of memory and takes too long to
process larger documents.
I'm preferring a cleaner and more lightweight solution, I'm not porting any
obsolete method from the OpenNLP, also I will improve the recognition of
the Portuguese in the library
In my mind, I'm committed to do this project... So, I prefer to do the best
I can, let's see what can be accomplished.
The good thing is that I'm reviewing every line of code OpenNLP, I’m taking
note of a few things that may not be correct, will soon share those notes
with the devs here.
Sounds good.
Do you also want to port all the training code, or only the code that is
necessary to run a model?
If you have any question about the data formats just ask here. We maybe
should also work
on documenting them.
We have some features that are Java only e.g. loading a user implemented
Java class, not sure if
that can be supported at all. Anyway, that is not a very common use case.
Jörn
