Check out this Cluster Analysis WhizzML script "Find neighbors"

Find neighbors | BigML.com.au

FREE

Find neighbors

Details:

Created	Wed, 8 Jun 2016 19:23:42 +0000
Published	Thu, 9 Jun 2016 08:10:27 +0000

Description:

Find the closest cluster rows to a given one

Tags:

Clusters Nearest neighbor

URL:

https://bigml.com.au/user/whizzml/gallery/script/575870be6501993439000068

Script FREE

Source code

;; Given a cluster and a point, find the closet n
;; rows in the cluster's dataset(s).

;; Given a cluster and an instance with the same fields as its dataset,
;; order the cluster's centroids by increasing distance to the instance.
(define (ordered-centroids cluster instance)
  (let (descs (get-in cluster ["clusters" "fields"])
        weights (get cluster "scales")
        cs (map (lambda (cn)
                  (let (c (get cn "center")
                        d (row-distance-squared descs weights c instance))
                  (assoc c "distance" (or d -1) "id" (get cn "id"))))
                (get-in cluster ["clusters" "clusters"]))
        cs (filter (lambda (c) (positive? (get c "distance"))) cs))
    (sort-by-key "distance" cs)))

;; Auxiliary function for error signaling.
(define (raise-missing id)
  (raise {"message" (str "Missing input field: " id) "code" -1}))

;; Auxiliary function: constructs the flatline string that generates a
;; new field with the distance of each row to the given one.
(define (distance-flatline cluster instance)
  (let (ids (get cluster "input_fields")
        ps (map (lambda (id) (or (get instance id) (raise-missing id))) ids)
        scales (get cluster "scales")
        ws (map (lambda (id) (get scales id 1)) ids))
    (flatline "(row-distance-squared (list @{{ps}}) (all) (list @{{ws}}))")))

;; Given a cluster and one of its centroids, uses the flatline
;; string generated by `distance-flatline` to create a new
;; dataset that extend's the centroid dataset with a distance
;; column.
(define (generate-distance-dataset cluster cent fl)
  (let (cluster-id (get cluster "resource")
        id (or (get cent "id") (raise (str "No id in " cent)))
        ds-id (get-in cluster ["cluster_datasets" id])
        ds-id (if (or (not ds-id) (empty? ds-id))
                  (create-and-wait-dataset {"cluster" cluster-id "centroid" id})
                  (str "dataset/" ds-id)))
    (create-and-wait-dataset {"origin_dataset" ds-id
                              "new_fields" [{"name" "distance" "field" fl}]})))

;; Given an extended centroid dataset (created by
;; `generate-distance-dataset`), returns the list of `n` rows with the
;; smallest values in the distance column.
(define (fetch-dataset-instances ds-id n)
  (let (sample-id (create-and-wait-sample {"dataset" ds-id})
        obj-id (dataset-get-objective-id ds-id)
        sample (fetch sample-id {"row_order_by" obj-id
                                 "rows" n
                                 "mode" "linear"})
        rows (get-in sample ["sample" "rows"] []))
    (delete sample-id)
    rows))

;; Final workflow.
(define (find-neighbors cluster-id instance n)
  (let (cluster (fetch cluster-id {"limit" -1})
        fl (distance-flatline cluster instance))
    (loop (cps (ordered-centroids cluster instance) instances [] m n)
      (cond (empty? cps) instances
            (< m 1) instances
            (let (ds-id (generate-distance-dataset cluster (head cps) fl)
                  new-instances (fetch-dataset-instances ds-id m))
              (recur (tail cps)
                     (concat instances new-instances)
                     (- m (count new-instances))))))))

;; Inputs and outputs
(define rows (find-neighbors cluster-id instance n))

Description

Inputs

Outputs

License: Creative Commons Zero V1.0
Script FREE

Find neighbors | BigML.com.au

Sending Request...

COMPANY

PRODUCT

BUSINESS

TRAINING

GALLERY

License

Embed this Script in your web site

COMPANY

PRODUCT

BUSINESS

TRAINING

GALLERY