I also gave it a try with he BioNLP data set we now support,
and I believe it is a really nice addition to the evaluators.
Yesterday I did not at all had an idea what kind of things are
frequently misclassified. Today I can simply add -printErros true
to the args and get a trace.
Jörn
On 7/13/11 9:58 AM, Jörn Kottmann wrote:
Hi,
for me it looks like that this error information can
also be important for other tools, e.g. ui tools or computing error
statistics and should be reported through an API back to the caller of
the
evaluator.
When instantiating an evaluator a user could register an interface
which is used to report the errors, the command line tools can then
implement this interface to print them out.
What do you think?
Jörn
