In mhahsler/stream: Infrastructure for Data Stream Mining

pkg <- 'stream'

source("https://raw.githubusercontent.com/mhahsler/pkg_helpers/main/pkg_helpers.R")
pkg_title(pkg)

Introduction

The package provides support for modeling and simulating data streams as well as an extensible framework for implementing, interfacing and experimenting with algorithms for various data stream mining tasks. The main advantage of stream is that it seamlessly integrates with the large existing infrastructure provided by R. The package provides:

Stream Sources: streaming from files, databases, in-memory data, URLs, pipes, socket connections and several data stream generators including dynamically streams with concept drift.
Stream Processing with filters (convolution, scaling, exponential moving average, ...)
Stream Aggregation: sampling, windowing.
Stream Clustering: BICO, BIRCH, D-Stream, DBSTREAM, and evoStream.
Stream Outlier Detection based on D-Stream, DBSTREAM.
Stream Classification with DecisionStumps, HoeffdingTree, NaiveBayes and Ensembles (streamMOA via RMOA).
Stream Regression with Perceptron, FIMTDD, ORTO, ... (streamMOA via RMOA).
Stream Mining Evaluation with prequential error estimation.

Additional packages in the stream family are:

streamMOA: Interface to clustering algorithms implemented in the MOA framework. The package interfaces clustering algorithms like of DenStream, ClusTree, CluStream and MCOD. The package also provides an interface to RMOA for MOA's stream classifiers and stream regression models.
rEMM: Provides implementations of threshold nearest neighbor clustering (tNN) and Extensible Markov Model (EMM) for modelling temporal relationships between clusters.

pkg_citation(pkg, 2)
pkg_install(pkg)

Usage

options(digits = 3)

Load the package and a random data stream with 3 Gaussian clusters and 10\% noise and scale the data to z-scores.

library("stream")
set.seed(2000)

stream <- DSD_Gaussians(k = 3, d = 2, noise= .1) %>% DSF_Scale()
get_points(stream, n = 5)

plot(stream)

Cluster a stream of 1000 points using D-Stream which estimates point density in grid cells.

dsc <- DSC_DStream(gridsize = .1)
update(dsc, stream, 1000)
plot(dsc, stream, grid = TRUE)

evaluate_static(dsc, stream, n = 100)

Outlier detection using DBSTREAM which uses micro-clusters with a given radius.

dso <- DSOutlier_DBSTREAM(r = .1)
update(dso, stream, 1000)
plot(dso, stream)

evaluate_static(dso, stream, n = 100, measure = c("numPoints", "noiseActual", "noisePredicted", "noisePrecision"))

Preparing complete stream process pipelines that can be run using a single update() call.

pipeline <- DSD_Gaussians(k = 3, d = 2, noise= .1) %>% 
  DSF_Scale() %>% 
  DST_Runner(DSC_DStream(gridsize = .1))
pipeline

update(pipeline, n = 500)
pipeline$dst