Thanks for this example, Mike. xdmp:plan is much easier to understand in ML7.
Now that result counts are correct, it’s more obvious that the Search API facet counts are often off by a few, always overcounting compared to the total returned after the search is executed with the related constraint. The problem seems to be that while cts:search is able to estimate result counts within only the fragments defined in the searchable expression, cts:element-values()/cts:frequency() does not. Therefore any ancestor document <chapter> of our fragment root <doc> will be added in with the facet estimate, while they are excluded from the search estimate. Is there a workaround, or is this just a pathological condition of using fragment roots? -Will On Nov 19, 2013, at 5:15 PM, Michael Blakeley <[email protected]> wrote: > That makes sense. For SEO purposes here's an example of how xdmp:plan might > help debug that sort of thing. The extra output in ML7 makes it clear that > with fast-phrase and without word-positions, only two-word terms are checked. > > It is also possible to figure this out from the ML6 plans, but I think the > new annotations make it easier to understand. > > -- Mike > > xdmp:plan( > cts:search(doc(), cts:word-query('dog cat rat'))) > > (: fast-phrase, no word-positions :) > <qry:query-plan xmlns:qry="http://marklogic.com/cts/query";> > <qry:info-trace>xdmp:eval("xdmp:plan(&#13;&#10; cts:search(doc(), > cts:word-query('dog cat ...", (), <options > xmlns="xdmp:eval"><database>14758162542116138691</database><modules>17366211626271...</options>)</qry:info-trace> > <qry:info-trace>Analyzing path for search: fn:doc()</qry:info-trace> > <qry:info-trace>Step 1 is searchable: fn:doc()</qry:info-trace> > <qry:info-trace>Path is fully searchable.</qry:info-trace> > <qry:info-trace>Gathering constraints.</qry:info-trace> > <qry:word-trace text="dog cat"> > <qry:key>2096356216808567173</qry:key> > </qry:word-trace> > <qry:word-trace text="cat rat"> > <qry:key>12758927055138826609</qry:key> > </qry:word-trace> > <qry:info-trace>Search query contributed 2 constraints: cts:word-query("dog > cat rat", ("lang=en"), 1)</qry:info-trace> > <qry:partial-plan> > <qry:term-query weight="1"> > <qry:key>2096356216808567173</qry:key> > <qry:annotation>pair(word("dog"),word("cat"))</qry:annotation> > </qry:term-query> > </qry:partial-plan> > <qry:partial-plan> > <qry:term-query weight="1"> > <qry:key>12758927055138826609</qry:key> > <qry:annotation>pair(word("cat"),word("rat"))</qry:annotation> > </qry:term-query> > </qry:partial-plan> > <qry:info-trace>Executing search.</qry:info-trace> > <qry:final-plan> > <qry:and-query> > <qry:term-query weight="1"> > <qry:key>2096356216808567173</qry:key> > <qry:annotation>pair(word("dog"),word("cat"))</qry:annotation> > </qry:term-query> > <qry:term-query weight="1"> > <qry:key>12758927055138826609</qry:key> > <qry:annotation>pair(word("cat"),word("rat"))</qry:annotation> > </qry:term-query> > </qry:and-query> > </qry:final-plan> > <qry:info-trace>Selected 0 fragments to filter</qry:info-trace> > <qry:result estimate="0"/> > </qry:query-plan> > > (: word-positions :) > <qry:query-plan xmlns:qry="http://marklogic.com/cts/query";> > <qry:info-trace>xdmp:eval("xdmp:plan(&#13;&#10; cts:search(doc(), > cts:word-query('dog cat ...", (), <options > xmlns="xdmp:eval"><database>18400529833056734238</database><root>/Users/mblakele/S...</options>)</qry:info-trace> > <qry:info-trace>Analyzing path for search: fn:doc()</qry:info-trace> > <qry:info-trace>Step 1 is searchable: fn:doc()</qry:info-trace> > <qry:info-trace>Path is fully searchable.</qry:info-trace> > <qry:info-trace>Gathering constraints.</qry:info-trace> > <qry:word-trace text="dog"> > <qry:key>5166487143365525844</qry:key> > </qry:word-trace> > <qry:word-trace text="cat"> > <qry:key>12545744176132597186</qry:key> > </qry:word-trace> > <qry:word-trace text="rat"> > <qry:key>12285550591485045727</qry:key> > </qry:word-trace> > <qry:info-trace>Search query contributed 1 constraint: cts:word-query("dog > cat rat", ("lang=en"), 1)</qry:info-trace> > <qry:partial-plan> > <qry:word-query weight="1" min-occurs="1" max-occurs="4294967295"> > <qry:KP pos="0"> > <qry:key>5166487143365525844</qry:key> > <qry:annotation>word("dog")</qry:annotation> > </qry:KP> > <qry:KP pos="1"> > <qry:key>12545744176132597186</qry:key> > <qry:annotation>word("cat")</qry:annotation> > </qry:KP> > <qry:KP pos="2"> > <qry:key>12285550591485045727</qry:key> > <qry:annotation>word("rat")</qry:annotation> > </qry:KP> > </qry:word-query> > </qry:partial-plan> > <qry:info-trace>Executing search.</qry:info-trace> > <qry:final-plan> > <qry:and-query> > <qry:word-query weight="1" min-occurs="1" max-occurs="4294967295"> > <qry:KP pos="0"> > <qry:key>5166487143365525844</qry:key> > <qry:annotation>word("dog")</qry:annotation> > </qry:KP> > <qry:KP pos="1"> > <qry:key>12545744176132597186</qry:key> > <qry:annotation>word("cat")</qry:annotation> > </qry:KP> > <qry:KP pos="2"> > <qry:key>12285550591485045727</qry:key> > <qry:annotation>word("rat")</qry:annotation> > </qry:KP> > </qry:word-query> > </qry:and-query> > </qry:final-plan> > <qry:info-trace>Selected 0 fragments to filter</qry:info-trace> > <qry:result estimate="0"/> > </qry:query-plan> > > On 19 Nov 2013, at 15:05 , Will Thompson <[email protected]> wrote: > >> I narrowed down the problem to 3+ word phrases. With that hunch, I enabled >> word positions, and after reindexing the estimates are now correct. >> >> I was thinking, incorrectly, that estimates would still be accurate with >> only fast phrase searches (and not word positions) enabled. But now that I >> look back at how that works, it’s clear that would only be true of 2-word >> phrases. >> >> -Will >> >> >> On Nov 19, 2013, at 3:23 PM, Michael Blakeley <[email protected]> wrote: >> >>> Which release is this? Is the problem limited to a particular word? If so, >>> what words? >>> >>> Have you tried a query trace or xdmp:plan yet? If you can run that with ML7 >>> that is even more useful. >>> >>> -- Mike >>> >>> On 19 Nov 2013, at 12:43 , Will Thompson <[email protected]> wrote: >>> >>>> I’m trying to determine why some search result estimates are overcounted. >>>> Documents generally look like: >>>> >>>> <chapter> >>>> <subchapter> >>>> <doc> >>>> <section> >>>> >>>> Fragment root is set on <doc> (and no ancestors or descendants of <doc>). >>>> count(//doc) = xdmp:estimate(//doc) => true. The searchable expression is >>>> xdmp:directory((‘dir1’, ‘dir2’, …), ‘infinity’)//doc. The word query >>>> specification explicitly includes <doc> and excludes document root. >>>> >>>> The documentation suggests to prevent overcounting we just ensure that 1) >>>> searchable expressions always select a fragment, and 2) there are no >>>> predicates applied to the searchable expression. Are there any other >>>> conditions that may cause overcounting of a simple word query? >>>> >>>> -Will >>>> _______________________________________________ >>>> General mailing list >>>> [email protected] >>>> http://developer.marklogic.com/mailman/listinfo/general >>>> >>> >>> _______________________________________________ >>> General mailing list >>> [email protected] >>> http://developer.marklogic.com/mailman/listinfo/general >>> >> >> _______________________________________________ >> General mailing list >> [email protected] >> http://developer.marklogic.com/mailman/listinfo/general >> > > _______________________________________________ > General mailing list > [email protected] > http://developer.marklogic.com/mailman/listinfo/general > _______________________________________________ General mailing list [email protected] http://developer.marklogic.com/mailman/listinfo/general
