Re: [Aspell-user] Reg. performance of suggestions

[Ananda Kumar C]

Kevin Atkinson wrote:

On Mon, 18 Jul 2005, Ananda Kumar C wrote:

  

We are trying to use aspell as the spell checker for our application and 
seeing a delay in getting the suggestions for a misspelled word to the 
tune of around 64 millisecs on an average. Checking a word for correct 
spelling takes only a few microsecs, which is very good. We could also 
get away with getting the best suggestion or the best 5 suggestions from 
the suggestion list. Is there a way to limit aspell (by tweaking the 
code, if need may be) to return only the best suggestion or the best 5 
suggestions, so that the time taken for getting the list of suggestions 
comes down?
    

Decreasing the number of suggestions will not decrees the overall time it 
takes to get them.

The config option "sug-mode" can be used to improve suggestion speed.  
"ultra" is the fasted mode.

Also, as a general rule, Aspell suggestion speed improves over time.  Thus 
I suggestion you upgrade to the latest version of Aspell 0.60.3, as that 
version made some significant improvements is suggestion time.

  

Hi Kevin,

Thanks a lot for your input. The performance has grown now, but I have an interesting observation.
I had run my tests against both the 0.60.3 and 0.60.2 versions. My test program randomly picks up a word from a really huge list of words (around 1.2 M words) and tests the words for correct spelling.
If a word is misspelt, I get the suggestions. The sug-mode config is now set to ultra.I run this test until I get around 500 misspelt words.

The results are as follows :
+--------------------------+-----------------+------------------+
|                          | ver. 0.60.3     | ver. 0.60.2      |
+--------------------------+-----------------+------------------+
|Avg. time for checking    |    17 microsecs |    6 microsecs   |
|Avg. time for suggesting  |  2087 microsecs | 1029 microsecs   |
+--------------------------+-----------------+------------------+

Although the performance has come up really good, what do you think is the reason why 0.60.3 doesnt perform better than the previous version?
Also, the package for 0.60.3 is not available in FreeBSD ports. Any ideas on when it will be up there?

Thanks again,
Anand.