sis-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Peter Karich (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SIS-45) Performance improvement of queryByPointRadius
Date Thu, 19 Apr 2012 20:30:42 GMT

    [ https://issues.apache.org/jira/browse/SIS-45?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13257774#comment-13257774
] 

Peter Karich commented on SIS-45:
---------------------------------

Now I understand why the time is always the same ... because searchRegion is calculated wrong.
It should somehow depend on the radius which will again improve query time a lot!

In my own quadtree implementation (http://karussell.github.com/GraphHopper/) I have this:

{code}
   public BBox createBBox(double lat, double lon, double radiusInKm) {
        if (radiusInKm <= 0)
            throw new IllegalArgumentException("Distance cannot be 0 or negative! " + radiusInKm
+ " lat,lon:" + lat + "," + lon);

        // length of a circle at specified lat / dist
        double dLon = (360 / (calcCircumference(lat) / radiusInKm));

        // length of a circle is independent of the longitude
        double dLat = (360 / (CalcDistance.C / radiusInKm));

        // Now return bounding box in coordinates
        return new BBox(lat + dLat, lon - dLon, lat - dLat, lon + dLon);
    }
{code}

but it is not copy and pastable due to different bounding box definitions
                
> Performance improvement of queryByPointRadius
> ---------------------------------------------
>
>                 Key: SIS-45
>                 URL: https://issues.apache.org/jira/browse/SIS-45
>             Project: Spatial Information Systems
>          Issue Type: Improvement
>            Reporter: Peter Karich
>
> If I didn't make a mistake a search with the normal distance method took ~2.4s and with
this distance method it takes only about 0.7s:
> {code}
>   public static double getHaversineDistance(double fromLat, double fromLon, double toLat,
double toLon) {
>     double dLat = Math.toRadians(toLat - fromLat);
>         double dLon = Math.toRadians(toLon - fromLon);
>         double a = Math.sin(dLat / 2) * Math.sin(dLat / 2)
>                 + Math.cos(Math.toRadians(fromLat)) * Math.cos(Math.toRadians(toLat))
* Math.sin(dLon / 2) * Math.sin(dLon / 2);
>         return EARTH_RADIUS * 2 * Math.asin(Math.sqrt(a));
>   }
> {code}
> Also one should think about normalizing the distance before the search so that one does
not need the whole last line which should give further speed improvements.
> I'm still unsure why it takes roughly the same time in my example for 10km, 20km and
40kmm where at every step a lot more nodes are involved. Normally I would say the mode nodes
- the more comparisons it'll take and the slower it should get. But it doesn't. Probably I'm
measuring wrong?

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message