active-analytics.clustering.k-medoids — de.active-group/active-analytics 0.3.0

choose-initial-medoids-randomly clj

(choose-initial-medoids-randomly xs k)

Randomly chooses a set of medoids to start out with.

cluster clj

(cluster xs medoids dissimilarity-fn)

Groups the data points xs by assigning each to its nearest medoid.

contribution clj

(contribution candidate x medoids dissimilarity-fn)

Calculates the contribution of x to the change in gain when adding candidate as additional medoid.

dissimilarity clj

(dissimilarity x y dissimilarities)

Looks up the dissimilarity of two points x and y in a dissimilarity map.

dissimilarity-function clj

(dissimilarity-function xs distance-fn cache?)

Returns a function to calculate the dissimilarity between two data points. If cache? is true, a map of dissimilarities is built and the resulting function will be a closure that does lookups inside the map. Otherwise distance-fn is returned.

dissimilarity-map clj

(dissimilarity-map xs distance-fn)

Calculates a map of dissimilarities which is indexed by all combinations (pairs) of points.

dissimilarity-sum clj

(dissimilarity-sum x xs dissimilarity-fn)

Calculates the sum of dissimilarities of a point to all other points.

find-medoid clj

(find-medoid xs dissimilarity-fn)

Finds the medoid of a set of data points xs, that is, the point with the lowest sum of dissimilarities.

gain clj

(gain candidate xs medoids dissimilarity-fn)

Calculates the gain associated with the choice of a medoid candidate.

k-medoids clj

(k-medoids xs
           distance-fn
           k
           &
           {:keys [medoid-init-mode cache?]
            :or {medoid-init-mode :random cache? true}})

Performs k-medoid clustering on the data points xs, given the desired number of clusters k and a function distance-fn that computes the distance between two points.

Accepts the keyword arguments :medoid-init-mode and :cache?.

• :medoid-init-mode determines the way the initial set of medoids is created. The default value, :random, corresponds to random choice of initial medoids, while :pam utilizes the PAM BUILD algorithm to greedily choose k 'good' starting medoids.
• If :cache? is true (the default), a map of dissimilarities is built which is then used to perform lookups instead of distance recalculations.

min-medoid-distance clj

(min-medoid-distance x medoids dissimilarity-fn)

Calculates the distance to the closest medoid.

nearest-medoid clj

(nearest-medoid x medoids dissimilarity-fn)

Finds the medoid that is closest to a point x.

next-pam-medoid clj

(next-pam-medoid xs medoids dissimilarity-fn)

Determines the best additional medoid, that is, the point with maximal gain w.r.t. the currently chosen medoids.

pam-build clj

(pam-build xs dissimilarity-fn k)

Performs the PAM BUILD initialization. That is, this greedily finds an initial set of medoids by iteratively choosing points which minimize dissimilarity.

step clj

(step xs medoids dissimilarity-fn)

Calculates the next generation of medoids from the current one.