Jan Lisse wrote:
Hi,
i have posted a short article about UIMA on my blog.
http://twelve-programmers.blogspot.com/2008/03/introducing-apache-uima.html
Maybe you are interested in reading it to get some feedback.
Some comments on your blog - It looks like you did a deep dive into
UIMA! and have got your arms around it. You mentioned Spring - we have
a new component in UIMA (currently in the "sandbox") called UIMA-AS
which is intended to be a replacement for the CPM - providing a more
flexible and configurable way to scale out analytics. It (internally)
uses Spring to manage a lot of the multiprocessing and wiring together
of various components used in this scaleout.
The descriptors that describe components and type systems can be rather
verbose. I read in your blog that you didn't try using the Component
Descriptor Editor (an Eclipse plugin tool). It was designed to hide
the verbosity of the descriptor from the user, and also to do a lot of
up-front checking and validation.
Regarding the Collection Reader - the concept of this has been
generalized in a component called the CAS Multiplier. This is described
here:
http://incubator.apache.org/uima/downloads/releaseDocs/2.2.1-incubating/docs/html/tutorials_and_users_guides/tutorials_and_users_guides.html#ugr.tug.cm
This may not address the issue you are concerned with, though. I look
forward to seeing more about your polling collection reader.
We are trying to avoid reinventing the wheel where proven solutions are
available in the open source community (especially under the Apache
license); but perhaps we can do even better (with some extra help:-) ) !
Thanks for your nice blogging about UIMA!
Generally i really like UIMA
and think you've done a good job so far! To integrate UIMA in my
companies
environment i have written some code that i'm willing to give back to
the community if there is interest.
I have developed a polling Collection reader implementation for
example, that permanently watches a directory for incoming files.
Additionally i have written a small maven 2 plugin as a wrapper around
the jcasgen utility. This way
i don't have to checkin generated java source in the scm and the
source is generated by maven instead.
If you are interested in adding this to the code base just drop me a
line.
Sounds interesting. We have some maven efforts also to produce PEAR
files, and this might be complementary.
With kind regards,
Jan
