Package 'streamMOA' reference manual

Title:	Interface for MOA Stream Clustering Algorithms
Description:	Interface for data stream clustering algorithms implemented in the MOA (Massive Online Analysis) framework (Albert Bifet, Geoff Holmes, Richard Kirkby, Bernhard Pfahringer (2010). MOA: Massive Online Analysis, Journal of Machine Learning Research 11: 1601-1604).
Authors:	Michael Hahsler [aut, cre, cph] (ORCID: <https://orcid.org/0000-0003-2716-1405>), John Forrest [aut, cph], Matthew Bolanos [ctb], Matthias Carnein [ctb], Dalibor Krleža [ctb]
Maintainer:	Michael Hahsler <[email protected]>
License:	GPL-3
Version:	1.4.0
Built:	2026-10-13 23:39:42 UTC
Source:	https://github.com/mhahsler/streamMOA

BICO: Fast computation of k-means coresets in a data stream

Description

This is an interface to the MOA implementation of BICO. The original BICO implementation by Fichtenberger et al is also available as stream::DSC_BICO.

Usage

DSC_BICO_MOA(
  Cluster = 5,
  Dimensions,
  MaxClusterFeatures = 1000,
  Projections = 10,
  k = NULL,
  space = NULL,
  p = NULL
)
DSC_BICO_MOA(
  Cluster = 5,
  Dimensions,
  MaxClusterFeatures = 1000,
  Projections = 10,
  k = NULL,
  space = NULL,
  p = NULL
)

Arguments

Cluster

Number of desired centers.

Dimensions

Number of dimensions in the input stream; this must be specified in advance.

MaxClusterFeatures

Maximum number of cluster features in the coreset.

Projections

Number of random projections used for the nearest neighbor search.

k

Alias for Cluster.

space

Alias for MaxClusterFeatures.

p

Alias for Projections.

Details

BICO maintains a tree which is inspired by the clustering tree of BIRCH, a SIGMOD Test of Time award-winning clustering algorithm. Each node in the tree represents a subset of these points. Instead of storing all points as individual objects, only the number of points, the sum and the squared sum of the subset's points are stored as key features of each subset. Points are inserted into exactly one node.

Author(s)

Matthias Carnein

References

Hendrik Fichtenberger, Marc Gille, Melanie Schmidt, Chris Schwiegelshohn, Christian Sohler: BICO: BIRCH Meets Coresets for k-Means Clustering. ESA 2013: 481-492

Examples

# data with 3 clusters and 2 dimensions
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

# cluster with BICO
bico <- DSC_BICO_MOA(Cluster = 3, Dimensions = 2)
update(bico, stream, 100)
bico

# plot micro and macro-clusters
plot(bico, stream, type = "both")
# data with 3 clusters and 2 dimensions
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

# cluster with BICO
bico <- DSC_BICO_MOA(Cluster = 3, Dimensions = 2)
update(bico, stream, 100)
bico

# plot micro and macro-clusters
plot(bico, stream, type = "both")

CluStream Data Stream Clusterer

Description

Implements the CluStream algorithm for data streams (Aggarwal et al., 2003).

Usage

DSC_CluStream(m = 100, horizon = 1000, t = 2, k = 5)
DSC_CluStream(m = 100, horizon = 1000, t = 2, k = 5)

Arguments

m

Maximum number of micro-clusters.

horizon

Time window used by CluStream.

t

Maximum boundary factor used to decide whether a new point belongs to a micro-cluster. The boundary is t times the root mean square deviation from the micro-cluster center.

k

Number of macro-clusters produced by weighted k-means.

Details

This is an interface to the MOA implementation of CluStream.

If k is specified, then CluStream applies a weighted k-means algorithm for reclustering (see Examples section below).

Value

An object of class DSC_CluStream (subclass of stream::DSC_Micro, DSC_MOA and stream::DSC).

Author(s)

Michael Hahsler and John Forrest

References

Aggarwal CC, Han J, Wang J, Yu PS (2003). "A Framework for Clustering Evolving Data Streams." In "Proceedings of the International Conference on Very Large Data Bases (VLDB '03)," pp. 81-92.

Bifet A, Holmes G, Pfahringer B, Kranen P, Kremer H, Jansen T, Seidl T (2010). MOA: Massive Online Analysis, a Framework for Stream Classification and Clustering. In Journal of Machine Learning Research (JMLR).

Examples

# data with 3 clusters and 5% noise
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = .05)

# cluster with CluStream
clustream <- DSC_CluStream(m = 50, horizon = 100, k = 3)
update(clustream, stream, 500)
clustream

plot(clustream, stream, type = "both")
# data with 3 clusters and 5% noise
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = .05)

# cluster with CluStream
clustream <- DSC_CluStream(m = 50, horizon = 100, k = 3)
update(clustream, stream, 500)
clustream

plot(clustream, stream, type = "both")

ClusTree Data Stream Clusterer

Description

Interface to the MOA implementation of the ClusTree data stream clustering algorithm (Kranen et al., 2009).

Usage

DSC_ClusTree(horizon = 1000, maxHeight = 8, lambda = NULL, k = NULL)
DSC_ClusTree(horizon = 1000, maxHeight = 8, lambda = NULL, k = NULL)

Arguments

horizon

Length of the time window.

maxHeight

Maximum height of the tree.

lambda

Value used to override the computed decay parameter.

k

If specified, use k-means with k clusters for reclustering.

Details

ClusTree uses a compact, self-adaptive index structure to maintain stream summaries. Kranen et al. (2009) suggest EM or k-means for reclustering.

Value

An object of class DSC_ClusTree (subclass of stream::DSC, DSC_MOA, stream::DSC_Micro).

Author(s)

Michael Hahsler and John Forrest

References

Philipp Kranen, Ira Assent, Corinna Baldauf, and Thomas Seidl. 2009. Self-Adaptive Anytime Stream Clustering. In Proceedings of the 2009 Ninth IEEE International Conference on Data Mining (ICDM '09). IEEE Computer Society, Washington, DC, USA, 249-258. doi:10.1109/ICDM.2009.47

Examples

# data with 3 clusters
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

clustree <- DSC_ClusTree(maxHeight = 3)
update(clustree, stream, 500)
clustree

plot(clustree, stream)

# Use the k-means reclusterer with k = 3 to create macro-clusters
clustree <- DSC_ClusTree(maxHeight = 3, k = 3)
update(clustree, stream, 500)
clustree

plot(clustree, stream, type = "both")
# data with 3 clusters
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

clustree <- DSC_ClusTree(maxHeight = 3)
update(clustree, stream, 500)
clustree

plot(clustree, stream)

# Use the k-means reclusterer with k = 3 to create macro-clusters
clustree <- DSC_ClusTree(maxHeight = 3, k = 3)
update(clustree, stream, 500)
clustree

plot(clustree, stream, type = "both")

DenStream Data Stream Clusterer

Description

Interface for the DenStream cluster algorithm for data streams implemented in MOA.

Usage

DSC_DenStream(
  epsilon,
  mu = 1,
  beta = 0.2,
  lambda = 0.001,
  initPoints = 100,
  offline = 2,
  processingSpeed = 1,
  recluster = TRUE,
  k = NULL
)
DSC_DenStream(
  epsilon,
  mu = 1,
  beta = 0.2,
  lambda = 0.001,
  initPoints = 100,
  offline = 2,
  processingSpeed = 1,
  recluster = TRUE,
  k = NULL
)

Arguments

epsilon

Maximum radius of a micro-cluster. Must be between 0 and 1.

mu

Minimum weight required for a core micro-cluster.

beta

Weight multiplier used to identify outlier micro-clusters. Must be between 0 and 1.

lambda

Decay constant.

initPoints

Number of points used to initialize the algorithm with DBSCAN.

offline

Multiplier applied to epsilon for reachability reclustering. Must be between 2 and 20.

processingSpeed

Number of incoming points per time unit, used for decay. Must be between 1 and 1000.

recluster

If TRUE, apply offline reachability reclustering.

k

If specified, choose a reachability threshold to produce this number of macro-clusters.

Details

DenStream reclusters micro-clusters using DBSCAN-style reachability. The threshold is epsilon * offline (with offline = 2 by default).

If k is specified, single-link hierarchical clustering chooses a reachability threshold that produces k macro-clusters.

Value

An object of class DSC_DenStream (subclass of stream::DSC, DSC_MOA, stream::DSC_Micro) or, for recluster = TRUE, an object of class stream::DSC_TwoStage.

Author(s)

Michael Hahsler and John Forrest

References

Cao F, Ester M, Qian W, Zhou A (2006). Density-Based Clustering over an Evolving Data Stream with Noise. In Proceedings of the 2006 SIAM International Conference on Data Mining, pp 326-337. SIAM.

Examples

# data with 3 clusters and 5% noise
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

# use Den-Stream with reachability reclustering
denstream <- DSC_DenStream(epsilon = .05)
update(denstream, stream, 500)
denstream

# plot macro-clusters
plot(denstream, stream, type = "both")

# plot micro-clusters
plot(denstream, stream, type = "micro")

# reclustering: Choose reclustering reachability threshold automatically to find 4 clusters
denstream2 <- DSC_DenStream(epsilon = .05, k = 4)
update(denstream2, stream, 500)
plot(denstream2, stream, type = "both")
# data with 3 clusters and 5% noise
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

# use Den-Stream with reachability reclustering
denstream <- DSC_DenStream(epsilon = .05)
update(denstream, stream, 500)
denstream

# plot macro-clusters
plot(denstream, stream, type = "both")

# plot micro-clusters
plot(denstream, stream, type = "micro")

# reclustering: Choose reclustering reachability threshold automatically to find 4 clusters
denstream2 <- DSC_DenStream(epsilon = .05, k = 4)
update(denstream2, stream, 500)
plot(denstream2, stream, type = "both")

D-Stream Data Stream Clustering Algorithm

Description

This is an interface to the MOA implementation of D-Stream. A C++ implementation (including reclustering with attraction) is available as stream::DSC_DStream.

Usage

DSC_DStream_MOA(decayFactor = 0.998, Cm = 3, Cl = 0.8, Beta = 0.3)
DSC_DStream_MOA(decayFactor = 0.998, Cm = 3, Cl = 0.8, Beta = 0.3)

Arguments

decayFactor

Decay factor applied to grid-cell density.

Cm

Threshold for classifying grid cells as dense.

Cl

Threshold for classifying grid cells as sparse.

Beta

Adjusts the window of protection for renaming previously deleted grids as sporadic.

Details

D-Stream creates an equally spaced grid and estimates the density in each grid cell using the count of points falling in the cells. Grid cells are classified based on density into dense, transitional and sporadic cells. The density is faded after every new point by a decay factor.

Notes

The implementation uses a 1-by-1 grid, so the example expands the data range.
The MOA implementation does not currently return micro-clusters.

Author(s)

Matthias Carnein

References

Yixin Chen and Li Tu. 2007. Density-based clustering for real-time stream data. In Proceedings of the 13th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '07). ACM, New York, NY, USA, 133-142.

Li Tu and Yixin Chen. 2009. Stream data clustering based on grid density and attraction. ACM Transactions on Knowledge Discovery from Data, 3(3), Article 12 (July 2009), 27 pages.

Examples

set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05, space_limit = c(0, 10))

# cluster with D-Stream
dstream <- DSC_DStream_MOA(Cm = 3)
update(dstream, stream, 1000)
dstream

# plot macro-clusters
plot(dstream, stream, type = "macro")
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05, space_limit = c(0, 10))

# cluster with D-Stream
dstream <- DSC_DStream_MOA(Cm = 3)
update(dstream, stream, 1000)
dstream

# plot macro-clusters
plot(dstream, stream, type = "macro")

Micro-cluster Continuous Outlier Detector (MCOD)

Description

Interface to the MOA implementation of the MCOD algorithm for distance-based data stream outlier detection.

Usage

DSC_MCOD(r = 0.1, t = 50, w = 1000, recheck_outliers = FALSE)

DSOutlier_MCOD(r = 0.1, t = 50, w = 1000, recheck_outliers = TRUE)

get_outlier_positions(x, ...)

recheck_outlier(x, outlier_correlated_id, ...)

clean_outliers(x, ...)
DSC_MCOD(r = 0.1, t = 50, w = 1000, recheck_outliers = FALSE)

DSOutlier_MCOD(r = 0.1, t = 50, w = 1000, recheck_outliers = TRUE)

get_outlier_positions(x, ...)

recheck_outlier(x, outlier_correlated_id, ...)

clean_outliers(x, ...)

Arguments

r

Radius used to search for neighbors.

t

Minimum number of neighbors required for a point not to be an outlier.

w

Sliding window width in data points.

recheck_outliers

If TRUE, allow detected outliers to be checked again.

x

A DSC_MCOD object.

...

Further arguments (currently ignored).

outlier_correlated_id

Identifier of the outlier to check again.

Details

The algorithm detects density-based outliers. An object $x$ is defined to be an outlier if there are less than $t$ objects lying at distance at most $r$ from $x$ .

Outliers are stored and can be retrieved with get_outlier_positions() and checked again with recheck_outlier().

Note: The implementation updates the clustering when predict() is called.

Value

An object of class DSC_MCOD (subclass of stream::DSC_Micro, DSC_MOA and stream::DSC).

Functions

get_outlier_positions(): Returns spatial positions of all current outliers.
recheck_outlier(): Re-check whether the outlier identified by outlier_correlated_id is still an outlier. Returns TRUE if it is.
clean_outliers(): Forget detected outliers (currently not implemented).

Author(s)

Dalibor Krleža

References

Kontaki M, Gounaris A, Papadopoulos AN, Tsichlas K, and Manolopoulos Y (2016). Efficient and flexible algorithms for monitoring distance-based outliers over data streams. Information Systems, Vol. 55, pp. 37-53. doi:10.1109/ICDE.2011.5767923

Examples

# Example 1: Clustering with MCOD
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)
mcod <- DSC_MCOD(r = .1, t = 3, w = 100)
update(mcod, stream, 100)
mcod

plot(mcod, stream, n = 100)

# Example 2: Predict outliers (have a class label of NA)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)
mcod <- DSOutlier_MCOD(r = .1, t = 3, w = 100)
update(mcod, stream, 100)

plot(mcod, stream, n = 100)

# Retrieve detected outlier positions.
get_outlier_positions(mcod)

# Example 3: evaluate on a stream
evaluate_static(mcod, stream, n = 100, type = "micro",
  measure = c("crand", "noisePrecision", "outlierjaccard"))
# Example 1: Clustering with MCOD
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)
mcod <- DSC_MCOD(r = .1, t = 3, w = 100)
update(mcod, stream, 100)
mcod

plot(mcod, stream, n = 100)

# Example 2: Predict outliers (have a class label of NA)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)
mcod <- DSOutlier_MCOD(r = .1, t = 3, w = 100)
update(mcod, stream, 100)

plot(mcod, stream, n = 100)

# Retrieve detected outlier positions.
get_outlier_positions(mcod)

# Example 3: evaluate on a stream
evaluate_static(mcod, stream, n = 100, type = "micro",
  measure = c("crand", "noisePrecision", "outlierjaccard"))

DSC_MOA Class

Description

An abstract class that inherits from the base class stream::DSC and provides the common functions needed to interface MOA clusterers.

Usage

DSC_MOA(...)
DSC_MOA(...)

Arguments

...

Further arguments (currently ignored).

Details

DSC_MOA is a subclass of stream::DSC for MOA-based clusterers. DSC_MOA classes operate in a different way in that the centers of the micro-clusters have to be extracted from the underlying Java object. This is done by using rJava to perform method calls directly in the JRI and converting the multi-dimensional Java array into a local R data type.

Note: The formula interface is currently not implemented for MOA-based clusterers. Use stream::DSF to select features instead.

Author(s)

Michael Hahsler and John Forrest

References

Albert Bifet, Geoff Holmes, Richard Kirkby, Bernhard Pfahringer (2010). MOA: Massive Online Analysis, Journal of Machine Learning Research 11: 1601-1604

streamKM++

Description

This is an interface to the MOA implementation of streamKM++.

Usage

DSC_StreamKM(sizeCoreset = 10000, numClusters = 5, length = 100000L, ...)
DSC_StreamKM(sizeCoreset = 10000, numClusters = 5, length = 100000L, ...)

Arguments

sizeCoreset

Size of the coreset.

numClusters

Number of clusters to compute.

length

Maximum number of points in the data stream.

...

Further arguments (currently ignored).

Details

streamKM++ uses a tree-based sampling strategy to build a small weighted sample of the stream (a coreset). The MOA implementation applies k-means++ to find the requested number of centers in the coreset.

Notes

The clusterer can process at most length points. Processing more points causes an ArrayIndexOutOfBoundsException.
The coreset is not exposed as micro-clusters; only macro-clusters can be requested.

Author(s)

Matthias Carnein

References

Marcel R. Ackermann, Christiane Lammersen, Marcus Maertens, Christoph Raupach, Christian Sohler, Kamil Swierkot. StreamKM++: A Clustering Algorithm for Data Streams. In: Proceedings of the 12th Workshop on Algorithm Engineering and Experiments (ALENEX '10), 2010.

Examples

set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

# cluster with streamKM++
streamkm <- DSC_StreamKM(sizeCoreset = 100, numClusters = 3, length = 1000)
update(streamkm, stream, 100)
streamkm

# plot macro-clusters (no access to micro-clusters)
plot(streamkm, stream)
set.seed(1000)
stream <- DSD_Gaussians(k = 3, d = 2, noise = 0.05)

# cluster with streamKM++
streamkm <- DSC_StreamKM(sizeCoreset = 100, numClusters = 3, length = 1000)
update(streamkm, stream, 100)
streamkm

# plot macro-clusters (no access to micro-clusters)
plot(streamkm, stream)

Base class for MOA-based data stream generators

Description

Abstract base class for MOA-based data stream generators. It inherits directly from stream::DSD.

Usage

DSD_MOA(...)
DSD_MOA(...)

Arguments

...

Further arguments (currently ignored).

Value

The abstract class cannot be instantiated and produces an error.

Author(s)

Michael Hahsler

References

Bifet A, Holmes G, Pfahringer B, Kranen P, Kremer H, Jansen T, and Seidl T (2010). MOA: Massive Online Analysis, a Framework for Stream Classification and Clustering. Journal of Machine Learning Research, 11, 1601–1604.

Examples

## Not run: 
DSD_MOA()

## End(Not run)
## Not run: 
DSD_MOA()

## End(Not run)

Random RBF Generator Events Data Stream Generator

Description

Generates random data using MOA's RandomRBFGeneratorEvents stream.

Usage

DSD_RandomRBFGeneratorEvents(
  k = 3,
  d = 2,
  numClusterRange = 3L,
  kernelRadius = 0.07,
  kernelRadiusRange = 0,
  densityRange = 0,
  speed = 100L,
  speedRange = 0L,
  noiseLevel = 0.1,
  noiseInCluster = FALSE,
  eventFrequency = 30000L,
  eventMergeSplitOption = FALSE,
  eventDeleteCreate = FALSE,
  modelSeed = NULL,
  instanceSeed = NULL
)
DSD_RandomRBFGeneratorEvents(
  k = 3,
  d = 2,
  numClusterRange = 3L,
  kernelRadius = 0.07,
  kernelRadiusRange = 0,
  densityRange = 0,
  speed = 100L,
  speedRange = 0L,
  noiseLevel = 0.1,
  noiseInCluster = FALSE,
  eventFrequency = 30000L,
  eventMergeSplitOption = FALSE,
  eventDeleteCreate = FALSE,
  modelSeed = NULL,
  instanceSeed = NULL
)

Arguments

k

Average number of centroids in the model.

d

Number of dimensions in the generated stream.

numClusterRange

Range for the number of clusters.

kernelRadius

Average radius of the micro-clusters.

kernelRadiusRange

Range of variation in micro-cluster radii.

densityRange

Range of variation in cluster density.

speed

Number of points between kernel movements.

speedRange

Range of variation in kernel speed.

noiseLevel

Proportion of noise points.

noiseInCluster

If TRUE, allow noise points inside clusters.

eventFrequency

Number of points between concept-drift events.

eventMergeSplitOption

If TRUE, enable cluster merge and split events.

eventDeleteCreate

If TRUE, enable cluster deletion and creation events.

modelSeed

Random seed for the cluster model.

instanceSeed

Random seed for generated instances.

Details

Only a subset of the parameters supported by the underlying MOA generator is exposed. If modelSeed or instanceSeed is NULL, a seed is sampled from R's random-number generator. Set these arguments explicitly to reproduce a stream; call set.seed() to make the generated default seeds reproducible.

By default, the generator creates three clusters with concept drift. Cluster locations move over time, and clusters may merge.

Value

An object of class DSD_RandomRBFGeneratorEvents (subclass of DSD_MOA, stream::DSD).

Author(s)

Michael Hahsler and John Forrest

References

Albert Bifet, Geoff Holmes, Bernhard Pfahringer, Philipp Kranen, Hardy Kremer, Timm Jansen, Thomas Seidl. MOA: Massive Online Analysis, a Framework for Stream Classification and Clustering Journal of Machine Learning Research (JMLR), 2010.

Examples

stream <- DSD_RandomRBFGeneratorEvents()
get_points(stream, 10)

if (interactive()) {
animate_data(stream, n = 5000, horizon = 100, xlim = c(0, 1), ylim = c(0, 1))
}
stream <- DSD_RandomRBFGeneratorEvents()
get_points(stream, 10)

if (interactive()) {
animate_data(stream, n = 5000, horizon = 100, xlim = c(0, 1), ylim = c(0, 1))
}

Package 'streamMOA'

Help Index

BICO: Fast computation of k-means coresets in a data stream

Description

Usage

Arguments

Details

Author(s)

References

See Also

Examples

CluStream Data Stream Clusterer

Description

Usage

Arguments

Details

Value

Author(s)

References

See Also

Examples

ClusTree Data Stream Clusterer

Description

Usage

Arguments

Details

Value

Author(s)

References

See Also

Examples

DenStream Data Stream Clusterer

Description

Usage

Arguments

Details

Value

Author(s)

References

See Also

Examples

D-Stream Data Stream Clustering Algorithm

Description

Usage

Arguments

Details

Author(s)

References

See Also

Examples

Micro-cluster Continuous Outlier Detector (MCOD)

Description

Usage

Arguments

Details

Value

Functions

Author(s)

References

See Also

Examples

DSC_MOA Class

Description

Usage

Arguments

Details

Author(s)

References

See Also

streamKM++

Description

Usage

Arguments

Details

Author(s)

References

See Also

Examples

Base class for MOA-based data stream generators

Description