Create a New Streaming K-Means Model

Initialises a streaming k-means model. The first k observations passed to kmeans_stream_update() will be used as initial centroids (warm-up phase). Clustering begins properly after k observations have been seen.

Usage

kmeans_stream_new(k, d, halflife = -1)

Arguments

k

a positive integer, number of clusters

d

a positive integer, number of features (must match observations passed to update/predict)

halflife

numeric learning rate. Use -1 (default) for count-based updates or a value in (0, 1) for fixed-rate updates (forgetful, recent points weighted more heavily)

Value

a list with components:

ptr

external pointer to the C model struct

k

number of clusters

d

number of features

halflife

learning rate mode

References

MacQueen, J. (1967). Some methods for classification and analysis of #' multivariate observations. Proceedings of the Fifth Berkeley Symposium, 1, 281-297.

Sculley, D. (2010). Web-scale k-means clustering. Proceedings of the 19th International Conference on World Wide Web, 1177-1178.

See also

kmeans_stream_update, kmeans_stream_centers, kmeans_batch

Examples

model <- kmeans_stream_new(k = 3L, d = 2L)
model$k
#> [1] 3
model$d
#> [1] 2