As many people have pointed out the benefits of InkAnalysis, we've been
doing some experimenting with it. In general it's working well for us,
but we've noted it is rather slow when we give it a hint node with a
wordlist factoid containing about 170,000 possible words. Obviously this
is a bit of an extreme, but it seems to be significantly slower that
we'd like and possibly slower than just using a recognition context.

Are there any tips for making this sort of recognition take place faster?

Have you tried it with RecognizerContext to see if it's just the size of the
word list or actually IA that's not performing well?

Also, is the hint node relatively small? Or is it a global hint?

--
Josh Einstein (Tablet PC MVP)
Einstein Technologies
Tablet Enhancements for Outlook - Try it free: www.tabletoutlook.com


"Chris DiPierro" <cdipierr@mi-corporation.com> wrote in message
news:OPR%23DqbEIHA.4712@TK2MSFTNGP04.phx.gbl...
> As many people have pointed out the benefits of InkAnalysis, we've been
> doing some experimenting with it. In general it's working well for us, but
> we've noted it is rather slow when we give it a hint node with a wordlist
> factoid containing about 170,000 possible words. Obviously this is a bit
> of an extreme, but it seems to be significantly slower that we'd like and
> possibly slower than just using a recognition context.
>
> Are there any tips for making this sort of recognition take place faster?

Yes, I've since tried it with RecognizerContext again and it is
significantly faster.

I'm not sure as to the question of the "relatively small" hit node. All
we've done to create the hint is as follows:

AnalysisHintNode hint = null;
hint = analyzer.CreateAnalysisHint();
hint.Location.MakeInfinite();
hint.Factoid = Factoid.WordList;
hint.SetWordList(<an array of strings>)
hint.WordMode = true;
hint.CoerceToFactoid = true;

I'm not familiar with the terminology of a "global hint".



Josh Einstein wrote:
> Have you tried it with RecognizerContext to see if it's just the size of
> the word list or actually IA that's not performing well?
>
> Also, is the hint node relatively small? Or is it a global hint?
>

Okay, that's a global hint. Try to make the location match the area in which
you're writing the word. Also, the other option is that you could use Ink
Analysis for parsing the ink into its writing structure and then defer the
recognition to a RecognizerContext object.

Also, are you using background recognition or the synchronous Analyze
method?

--
Josh Einstein (Tablet PC MVP)
Einstein Technologies
Tablet Enhancements for Outlook - Try it free: www.tabletoutlook.com


"Chris DiPierro" <cdipierr@mi-corporation.com> wrote in message
news:471CA9CE.4020406@mi-corporation.com...
> Yes, I've since tried it with RecognizerContext again and it is
> significantly faster.
>
> I'm not sure as to the question of the "relatively small" hit node. All
> we've done to create the hint is as follows:
>
> AnalysisHintNode hint = null;
> hint = analyzer.CreateAnalysisHint();
> hint.Location.MakeInfinite();
> hint.Factoid = Factoid.WordList;
> hint.SetWordList(<an array of strings>)
> hint.WordMode = true; hint.CoerceToFactoid = true;
>
> I'm not familiar with the terminology of a "global hint".
>
>
>
> Josh Einstein wrote:
>> Have you tried it with RecognizerContext to see if it's just the size of
>> the word list or actually IA that's not performing well?
>>
>> Also, is the hint node relatively small? Or is it a global hint?
>>

Hi Chris,

There does seem to be a perf issue with large wordlists which we are
investigating. In the meantime, if you need to use such a large wordlist for
recognition, a RecognizerContext object would be much faster. You can use
the IntermediateResultsUpdated event to catch the Analyze() process before it
starts the recognition pass.

Thanks,
Amber P. Race [MS]


"Chris DiPierro" wrote:

> Yes, I've since tried it with RecognizerContext again and it is
> significantly faster.
>
> I'm not sure as to the question of the "relatively small" hit node. All
> we've done to create the hint is as follows:
>
> AnalysisHintNode hint = null;
> hint = analyzer.CreateAnalysisHint();
> hint.Location.MakeInfinite();
> hint.Factoid = Factoid.WordList;
> hint.SetWordList(<an array of strings>)
> hint.WordMode = true;
> hint.CoerceToFactoid = true;
>
> I'm not familiar with the terminology of a "global hint".
>
>
>
> Josh Einstein wrote:
> > Have you tried it with RecognizerContext to see if it's just the size of
> > the word list or actually IA that's not performing well?
> >
> > Also, is the hint node relatively small? Or is it a global hint?
> >
>

Amber --

Thanks for this. We've resorted to RecognizerContext for now which is
serving us fine. We'll investigate Analyzer again in a later release.


Amber P. Race wrote:
> Hi Chris,
>
> There does seem to be a perf issue with large wordlists which we are
> investigating. In the meantime, if you need to use such a large wordlist for
> recognition, a RecognizerContext object would be much faster. You can use
> the IntermediateResultsUpdated event to catch the Analyze() process before it
> starts the recognition pass.
>
> Thanks,
> Amber P. Race [MS]
>
>
> "Chris DiPierro" wrote:
>
>> Yes, I've since tried it with RecognizerContext again and it is
>> significantly faster.
>>
>> I'm not sure as to the question of the "relatively small" hit node. All
>> we've done to create the hint is as follows:
>>
>> AnalysisHintNode hint = null;
>> hint = analyzer.CreateAnalysisHint();
>> hint.Location.MakeInfinite();
>> hint.Factoid = Factoid.WordList;
>> hint.SetWordList(<an array of strings>)
>> hint.WordMode = true;
>> hint.CoerceToFactoid = true;
>>
>> I'm not familiar with the terminology of a "global hint".
>>
>>
>>
>> Josh Einstein wrote:
>>> Have you tried it with RecognizerContext to see if it's just the size of
>>> the word list or actually IA that's not performing well?
>>>
>>> Also, is the hint node relatively small? Or is it a global hint?
>>>