Hi, Thanks for the fast answer. That's also what I sort of figured out by searching on the web but it's good to know someone has implemented it :)
The "rough" positioning is a very good idea, thanks. I am pretty sure we have the kind of algorithm you are looking for but it is in a commercial product. Ping me off list if you are interested. Stéphane On Feb 17, 2008 12:43 PM, Max Metral <[EMAIL PROTECTED]> wrote: > We're doing this for our site (http://boston.povo.com) the simple way: have > Lucene return all matches based on non-geo criteria and then fetch the items > from the db by id and run our geo logic. We store some "rough" positioning > in Lucene, such as the region and use that for first level rejection. (If > your geometries can get artificially large then of course this doesn't fully > work, but in that case perhaps you store "total area" or some flag that says > this item spans a region. > > We briefly considered a bounding box model for Lucene to do better first > level rejection, but performance was more than good enough with the approach > I outlined above for our corpus. > > There's one thing we never implemented, which was calculating the minimum > distance between two geometries (we almost always have one side of the > comparison as a point). Do you happen to know a reasonably speedy algorithm > to do this? > > Thanks! > --Max > > > -----Original Message----- > From: Stephane Nicoll [mailto:[EMAIL PROTECTED] > Sent: Sunday, February 17, 2008 5:22 AM > To: java-user@lucene.apache.org > Subject: Using lucene with a Geospatial catalog > > Hi, > > I've been browsing the archive and the documentation about Lucene. It > really seems that it could help implementing my use case but I would > like to be sure first. > > What I need is to be able to search data in a "catalog" which is > geo-enabled. The data is stored in a database. A record has namely > > * name > * keywords > * footprint (that is a geometry that represents the record) > * date range (optional) that defines the "validity of the data" > * Timestamp > * Creation date > * various boolean flag > * custom data > > I understand that Lucene is powerful for full-text based search but > what about searching something like "give me all records that contains > the kewords foo, with flag bar true, valid between 20070602 and > 20070907 and whose geometry intersect a given box. > > I've seen on the list people using tricks by storing the coordinates > in a way we could use range. In my case, the geometry is potentially > very complex. The database handle that for me (Oracle Spatial or > PostGIS for instance) with intersect, contains and such. Is it > possible to combine a lucene search with a DB query? Any best > practices on that topic? > > Thanks, > Stéphane > > > -- > Large Systems Suck: This rule is 100% transitive. If you build one, > you suck" -- S.Yegge > > --------------------------------------------------------------------- > 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] > > -- Large Systems Suck: This rule is 100% transitive. If you build one, you suck" -- S.Yegge --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
