在mongodb中聚类地理数据

Question

We have a mongodb database with >200K rows each containing a point location (lat,lng). We would like to create a query that specifies a geo point & radius and return a list of clusters. Each cluster is basically an aggregation of locations that are near each other.

First question: Is it possible for mongodb to automatically create and maintain these clusters for us? and if yes, how can we query mongodb to return clusters (not actual data points) for a specific geo-location. Each returned cluster would have a position and the number of actual data points (geo-tagged rows). Basically, we would want it to return the equivalent of a k-means clustering algorithm.

We've created a mongodb geoHaystack index that seems to cluster rows but not sure how we can use it to achieve the above query:

db.locations.createIndex( { 'position' : "geoHaystack", type : 1 } , { bucketSize : 1 })

Alternatively, we could dynamically use a clustering algorithm such as https://github.com/spember/geo-cluster to generate these clusters but I'm assuming this would be a very slow process.

Any recommendations on how best to implement such a query?

Answer 1

In MongoDB, a geoHaystack index has another purpose - it is a special index that is optimized to return results over small areas. I think it can't be used here.

So, I think you can retrieve all points and do clustering using k-means. That should be fast. After that you can save them as another entities (eg Polygon) and use it anywhere you need.