"
This query supports up to 2 distances including transpositions,
anything beyond that is basically going to match a significant portion
of the term dictionary and not really be useful.
"
That would be a great statement to add to the Javadoc and address most of my
concern.
-- Jack Krupansky
-----Original Message-----
From: Robert Muir
Sent: Thursday, September 13, 2012 1:45 PM
To: [email protected]
Subject: Re: Javadoc issue: FuzzyQuery should note the editing distance
limit of 2 and refer to SlowFuzzyQuery
I dont agree with suggesting the slow, unscalable approach.
This query supports up to 2 distances including transpositions,
anything beyond that is basically going to match a significant portion
of the term dictionary and not really be useful.
If someone has special data where this makes sense, they should use an
n-gram indexing technique or the spellchecker module, or BLAST or
something other than Lucene.
As far as the constant of 2: this is actually in the javadocs: you
have to click CONSTANT VALUES.
On Thu, Sep 13, 2012 at 12:03 PM, Jack Krupansky
<[email protected]> wrote:
The automaton support for FuzzyQuery added the severe limitation to
FuzzyQuery of an editing distance of 2 that needs to be documented in the
Javadoc. A reference to SlowFuzzyQuery is also needed in the Javadoc, even
though that class is deprecated.
The constructor Javadoc does say “maxEdits - must be >= 0 and <=
LevenshteinAutomata.MAXIMUM_SUPPORTED_DISTANCE”, but neither the text nor
that link documents the extreme limitation of 2. I mean, a casual reader
might reasonably expect that it is just some big number like
Integer.MAX_VALUE. The rationale from the Jira should be succinctly
stated,
at the class level as well.
Relevant Jira:
https://issues.apache.org/jira/browse/LUCENE-4024
-- Jack Krupansky
--
lucidworks.com
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
