You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@uima.apache.org by Marshall Schor <ms...@schor.com> on 2006/11/25 17:11:52 UTC
Result specification - update needed
I need to write up the version 2 tutorial and user's guide for Results
Specification. The current write up is inaccurate, I think. I started
to change it to fit the new API where it is not passed in as a
parameter, but there are more things that need fixing.
Could Adam and/or Thilo take a look at this write up and fix it up?
(see below):
-Marshall
<section id="ugr.tug.aae.result_specification_setting">
<title>Result Specification Setting</title>
<para>The Result Specification is passed to the annotator instance
by calling its
setResultSpecificaiton method. When called, the default
implementation saves the
result specification in an instance variable of the Annotator
instance.</para>
<para>A results specification is a list of output types and / or
type:feature
specifications, which are expected to be <quote>output</quote>
from the
annotator. Annotators may use this to optimize their operations,
when possible, for
those cases where only particular outputs are wanted. The
interface to the Result
Specification object (see the JavaDocs) allows querying both
types and particular
features of types.</para>
<para>Sometimes you can specify the Result Specification;
othertimes, you cannot
(for instance, inside a Collection Processing Engine, you
cannot). When you cannot
specify it, or choose not to specify it (for example, using the
form of the
process(...) call on an Analysis Engine that doesn't
include the Result
Specification), a <quote>Default</quote> Result Specification is
used.</para>
</section>
<section><title>Default ResultSpecification</title>
<para>The default Result Specification is taken from the
Engine's output
Capability Specification. Remember that a Capability
Specification has both
inputs and outputs, can specify types and / or features, and
there can be more than one
Capability Set. If there is more than one set, the logical union
of these sets is used.
The default Result Specification is exactly what's included
in the output
Capability Specification.</para>
</section>
<section><title>Passing Result Specifications to Annotators</title>
<para>If you are not using aggregation or collection processing,
but instead are
instantiating your own primitive analysis engines and calling
their process
methods, you can pass whatever Result Specification is
appropriate in your call to
process(CAS, ResultSpecification). For primitive engines,
whatever you pass in is
passed along as the value of the 2nd argument in the
annotator's process()
method. If you use the form of the call without the Result
Specification, the default
Result Specification is created and passed, as above.</para>
</section>
<section><title>Aggregates</title>
<para>For aggregate engines, the value passed to the primitive
annotator code depends
on the kind of flow.</para>
</section>
<section><title>Fixed Flow</title>
<para>For FixedFlow, any ResultSpecification passed into the
aggregate is ignored,
and instead, each primitive annotator is passed a result spec
that corresponds to the
union of its output capability specifications at the primitive
descriptor level. If
no output capability specification is given, the annotator will
still be called, but
the result specification will be empty.</para>
</section>
<section><title>CapabilityLanguageFlow</title>
<para>For CapabilityLanguageFlow, each annotator is passed a
ResultSpecification
that is the intersection of the primitive annotator's
output Capability
Specification with the ResultSpecification passed to the
aggregate. If this
intersection is null (the annotator does not produce any type or
feature included in
the ResultSpecification), the annotator will not be called at
all.</para>
<para>Therefore, if using the CapabilityLanguageFlow, if you want
to supply a custom
ResultSpecification for the aggregate it must include any
intermediate types that
need to be produced internally in the flow, or else things will
not work
properly.</para>
</section>
<section><title>Special rule for skipping Analysis Engines</title>
<para>When using the CapabilityLanguageFlow, an annotator will be
also be skipped if
all of its outputs are in the output capability of some
annotator(s) that has (have)
executed previously in the flow. The concept here is that if all
of an
annotator's output types have already been produced, that
annotator will not
be called.</para>
<para>For an Aggregate, each annotator is passed a Result
Specification that is the
intersection of the set of types mentioned in its output with
the Result
Specification passed to the aggregate. If this intersection is
null (the annotator
does not produce any type included in the ResultSpecification),
the annotator will
not be called at all.</para>
<para>Therefore, if using the CapabilityLanguageFlow, if you want
to supply a custom
ResultSpecification for the aggregate it must include any
intermediate types that
need to be produced, or else things will not work properly.</para>
</section>
<section><title>Collection Proessing Engines</title>
<para>The Default Result Specification is always used for all
components of a
Collection Processing Engine.</para>
</section>
Re: Result specification - update needed
Posted by Adam Lally <al...@alum.rpi.edu>.
On 11/25/06, Marshall Schor <ms...@schor.com> wrote:
> I need to write up the version 2 tutorial and user's guide for Results
> Specification. The current write up is inaccurate, I think. I started
> to change it to fit the new API where it is not passed in as a
> parameter, but there are more things that need fixing.
>
> Could Adam and/or Thilo take a look at this write up and fix it up?
> (see below):
> <snip/>
Yes, this needed an overhaul. Result Specifcation handling in
aggregates no longer has anything to do with the type of flow. Here's
my suggested documentation (note I used <code/> tags for monospace
font as in HTML, I have no idea if that's right for docbook):
<section id="ugr.tug.aae.result_specification_setting">
<title>Result Specification Setting</title>
<para>The Result Specification is passed to the annotator instance by
calling its
setResultSpecificaiton method. When called, the default
implementation saves the
result specification in an instance variable of the Annotator
instance, which can be
accessed by the annotator using the protected
<code>getResultSpecification()</code> method.</para>
<para>A Result Specification is a list of output types and / or type:feature
names, which are expected to be
<quote>output</quote> from the annotator. Annotators may use this to optimize
their operations, when possible, for those cases where only
particular outputs are
wanted. The interface to the Result Specification object (see the
JavaDocs) allows
querying both types and particular features of types.</para>
<para>Sometimes you can specify the Result Specification; othertimes,
you cannot (for
instance, inside a Collection Processing Engine, you cannot). When you cannot
specify it, or choose not to specify it (for example, using the form of the
process(...) call on an Analysis Engine that doesn't include the Result
Specification), a
<quote>Default</quote> Result Specification is used.</para>
</section>
<section>
<title>Default ResultSpecification</title>
<para>The default Result Specification is taken from the Engine's output
Capability Specification. Remember that a Capability Specification has both
inputs and outputs, can specify types and / or features, and there
can be more than one
Capability Set. If there is more than one set, the logical union of
these sets is used.
The default Result Specification is exactly what's included in the output
Capability Specification.</para>
</section>
<section>
<title>Passing Result Specifications to Analysis Engines</title>
<para>If you are not using a Collection Processing Engine, you can
specify a Result Specification
for your AnalysisEngine(s) by calling the
<code>AnalysisEngine.setResultSpecification(ResultSpecification)</code>
method.</para>
<para>It is also possible to pass a Result Specification on each call to
<code>AnalysisEngine.process(CAS, ResultSpecification)</code>.
However, this is not recommended
if your Result Specification will stay constant across multiple
calls to <code>process</code>.
In that case it will be more efficient to call
<code>AnalysisEngine.setResultSpecification(ResultSpecification)</code>
only when the Result Specification changes.</para>
<para>
For primitive Analysis Engines, whatever Result Specification you pass in is
passed along to the annotator's
<code>setResultSpecification(ResultSpecification)</code>
method. For aggregate Analysis Engines, see below.</para>
</section>
<section>
<title>Aggregates</title>
<para>For aggregate engines, the Result Specification passed to the
<code>AnalysisEngine.setResultSpecification(ResultSpecification)</code>
method is intended
to specify the set of output types/features that the aggregate
should produce. This is not
necessarily equivalent to the set of output types/features that each
annotator should produce.
For example, an annotator may need to produce an intermediate type
that is then consumed
by a downstream annotator, even though that intermediate type is not
part of the Result
Specification.</para>
<para>To handle this situation, when
<code>AnalysisEngine.setResultSpecification(ResultSpecification)</code>
is called on
an aggregate, the framework computes the union of the passed Result
Specification with the set of
<emph>all</emph> input types and features of <emph>all</emph>
component AnalysisEngines within that
aggregate. This forms the complete set of types and features that
any component of the aggregate
might need to produce. This derived Result Specification is then
passed to the
<code>AnalysisEngine.setResultSpecification(ResultSpecification)</code>
of each component AnalysisEngine.
In the case of nested aggregates, this procedure is applied
recursively.</para>
</section>
<section>
<title>Collection Proessing Engines</title>
<para>The Default Result Specification is always used for all
components of a Collection
Processing Engine.</para>
</section>
<!--
This no longer belongs as part of the discussion of Rsult Specifications.
The CapabilityLanguageFlow now skips annotators on the basis of their complete
capabilities, it does not take the Result Specification into account.
Result Specifications are no longer the concern of the Flow
Controller, since this
was deemed to be too great a complexity without enough benefit.
<section>
<title>Special rule for skipping Analysis Engines</title>
<para>When using the CapabilityLanguageFlow, an annotator will be
also be skipped if all
of its outputs are in the output capability of some annotator(s)
that has (have)
executed previously in the flow. The concept here is that if all of an
annotator's output types have already been produced, that
annotator will not
be called.</para>
<para>For an Aggregate, each annotator is passed a Result
Specification that is the
intersection of the set of types mentioned in its output with the Result
Specification passed to the aggregate. If this intersection is null
(the annotator
does not produce any type included in the ResultSpecification), the
annotator will
not be called at all.</para>
<para>Therefore, if using the CapabilityLanguageFlow, if you want to
supply a custom
ResultSpecification for the aggregate it must include any
intermediate types that
need to be produced, or else things will not work properly.</para>
</section>
-->