support geo sorting on multiple geo point values per doc

[tommyvn]

I have multiple geo points per document and would like to be able to sort based on the closest matching location in the document to my center point. Apparently the way things currently work is the sort will be based on a random location on the document, as in the following scenario:

I have the following mapping:

$ curl -XGET 'http://localhost:9200/work/offices/_mapping?pretty=true'
{
  "offices" : {
    "properties" : {
      "location" : {
        "properties" : {
          "address" : {
            "type" : "string"
          },
          "point" : {
            "type" : "geo_point"
          }
        }
      },
      "name" : {
        "type" : "string"
      }
    }
  }
}

Then I have the following query and response (I've only included hits to cut down on noise):

$ curl -XGET 'http://localhost:9200/work/offices/_search?pretty=true' -d '{ "fields": [ "name", "location" ], "query": { "match_all": {} }, "sort": { "_geo_distance": {"location.point": [-0.0976398, 51.4962307], "order": "asc", "unit": "km"} }}'
...
    "hits" : [ {
      "_index" : "work",
      "_type" : "offices",
      "_id" : "IDtt2WnSQnWEhuAxhlbPgw",
      "_score" : null,
      "fields" : {
        "location" : [ {
          "point" : [ -0.01655, 51.5007324 ],
          "address" : "E14 9SH"
        }, {
          "point" : [ -0.0976398, 51.4962307 ],
          "address" : "SE1 6PL"
        } ],
        "name" : "office3"
      },
      "sort" : [ 0.0 ]
    }, {
      "_index" : "work",
      "_type" : "offices",
      "_id" : "xzaVAHUoSGON8gY1-ggILQ",
      "_score" : null,
      "fields" : {
        "location" : [ {
          "point" : [ -0.01655, 51.5007324 ],
          "address" : "E14 9SH"
        }, {
          "point" : [ -0.0684337, 51.4843866 ],
          "address" : "SE1 5BA"
        } ],
        "name" : "office10"
      },
      "sort" : [ 2.413161036894697 ]
    }, {
      "_index" : "work",
      "_type" : "offices",
      "_id" : "ZWjuiVkWSD6H99ooKcKNuA",
      "_score" : null,
      "fields" : {
        "location" : [ {
          "point" : [ -0.0976398, 51.4962307 ],
          "address" : "SE1 6PL"
        }, {
          "point" : [ -0.0684337, 51.4843866 ],
          "address" : "SE1 5BA"
        } ],
        "name" : "office8"
      },
      "sort" : [ 2.413161036894697 ]
    }
...

office8 and office3 both have locations 0km away from my search point, yet office10 is sneaking in between the two at 2.4km away and office8 is also placing itself 2.4km away (which it is, but but only on the further location point).

This was originally discussed here: https://groups.google.com/forum/?fromgroups#!topic/elasticsearch/DxIUevwZfOs

[DominicWatson]

While waiting for this enhancement, I've managed to get this working using the scripting functionality of ElasticSearch (which I have to say is utterly awesome, this is such a refreshing API to work with).

The following solution is very rough. It currently only supports kilometers and the lang-javascript plugin is required.

A temporary solution

I have the following javascript in ./config/scripts/geo/closestdistance.js:

(function(){
    var i, locations, calculateDistance, closest, distance;

    if ( !doc['latlon'].multiValued ){
        return doc['latlon'].distanceInKm( lat, lon )
    }

    calculateDistance = function( lat1, lat2, lon1, lon2 ){
        var R = 6371 // km
          , dLat = (lat2-lat1) * Math.PI / 180
          , dLon = (lon2-lon1) * Math.PI / 180
          , lat1 = lat1 * Math.PI / 180
          , lat2 = lat2 * Math.PI / 180
          , a = Math.sin(dLat/2) * Math.sin(dLat/2) + Math.sin(dLon/2) * Math.sin(dLon/2) * Math.cos(lat1) * Math.cos(lat2)
          , c = 2 * Math.atan2(Math.sqrt(a), Math.sqrt(1-a));

        return R * c;
    }

    locations = doc['latlon'].getValues();
    closest = calculateDistance( lat, locations[0].lat(), lon, locations[0].lon() );

    for( i=1; i < locations.length; i++ ){
        distance = calculateDistance( lat, locations[i].lat(), lon, locations[i].lon() );
        closest = distance < closest ? distance : closest;
    }

    return closest;
})();

Then, in my search body, I have something like:

{
    query : {
        custom_score : {
              query : {
                  query_string : {
                      query : "myquerystring..."
                  }
              }
            , params : {
                  lat : 54.5881
                , lon : -5.85829
              }
            , script : "geo_closestdistance"
        }
    },
    sort : [{_score : { order : "asc"}}]
}

Hopefully someone will find this useful.

[ghost]

jest sewz you know dominic: ESP dev centre left geo-location searching
as a rectangulation problem, despite how the measurements are laid out
on the globe. That was the search engine bought by Microsoft. I've no
idea if they got geo into SharePoint search and whether they got
something smarter than right angles... i look forward to leveraging this
solution.

b.

On 09/07/2012 05:48 PM, Dominic Watson wrote:

While waiting for this enhancement, I've managed to get this working
using the scripting functionality of ElasticSearch (which I have to
say is utterly awesome, this is such a refreshing API to work with).

The following solution is very rough. It currently only supports
kilometers and the |lang-javascript| plugin is required.

A temporary solution

I have the following javascript in
|./config/scripts/geo/closestdistance.js|:

|(function(){
var i, locations, calculateDistance, closest, distance;

 if ( !doc['latlon'].multiValued ){
     return doc['latlon'].distanceInKm( lat, lon )
 }

 calculateDistance = function( lat1, lat2, lon1, lon2 ){
     var R = 6371 // km
       , dLat = (lat2-lat1) * Math.PI / 180
       , dLon = (lon2-lon1) * Math.PI / 180
       , lat1 = lat1 * Math.PI / 180
       , lat2 = lat2 * Math.PI / 180
       , a = Math.sin(dLat/2) * Math.sin(dLat/2) + Math.sin(dLon/2) * Math.sin(dLon/2) * Math.cos(lat1) * Math.cos(lat2)
       , c = 2 * Math.atan2(Math.sqrt(a), Math.sqrt(1-a));

     return R * c;
 }

 locations = doc['latlon'].getValues();
 closest = calculateDistance( lat, locations[0].lat(), lon, locations[0].lon() );

 for( i=1; i < locations.length; i++ ){
     distance = calculateDistance( lat, locations[i].lat(), lon, locations[i].lon() );
     closest = distance < closest ? distance : closest;
 }

 return closest;

})();
|

Then, in my search body, I have something like:

|{
query : {
custom_score : {
query : {
query_string : {
query : "myquerystring..."
}
}
, params : {
lat : 54.5881
, lon : -5.85829
}
, script : "geo_closestdistance"
}
},
sort : [{_score : { order : "asc"}}]
}
|

Hopefully someone will find this useful.

—
Reply to this email directly or view it on GitHub
#1846 (comment).

[tudborg]

I would also very much like to see this implemented.
Preferably controlled by the mode option as described in
http://www.elasticsearch.org/guide/reference/api/search/sort/

[tudborg]

As a response to @DominicWatson:
This will run ~30% faster:

(function(){
    var i, locations, arcDist, closest, distance;
    arcDist = function( lat1, lat2, lon1, lon2 ){
        var R = 6371 // earth radius
          , CI = 0.017453292519943295 // pi/180
          , dLat = (lat2-lat1) * CI / 2
          , dLon = (lon2-lon1) * CI / 2
          , sinDLat = Math.sin(dLat)
          , sinDLon = Math.sin(dLon)
          , a = sinDLat*sinDLat + sinDLon*sinDLon * Math.cos(lat1*CI)*Math.cos(lat2*CI)
          , c = 2*Math.atan2(Math.sqrt(a), Math.sqrt(1-a));
        return R * c;
    }
    locations = doc["locations"].getValues();
    closest = arcDist( lat, locations[0].lat(), lon, locations[0].lon() );
    for( i=1; i < locations.length; i++ ) {
        distance = arcDist( lat, locations[i].lat(), lon, locations[i].lon() );
        if(distance < closest) {
          closest = distance;
        }
    }
    return closest;
})();

[martijnvg]

The next 0.90 release will have a mode option (min & max) support for sorting by geo distance.

[mouzt]

Does this question have new solution which is not rough, now