In zedoul/cbar: Contextual Bayesian Anomaly Detection in R

library(cbar, warn.conflicts = F)
library(dplyr, warn.conflicts = F)
library(ggplot2, warn.conflicts = F)

Installation

Either you try stable CRAN version

install.packages("cbar")

Or unstable development version

devtools::install_github("zedoul/cbar")

You'll need to use library to load as follows:

library(cbar)

Introduction

cbar (Contextual Bayesian Anomaly Detection in R) is an R package for detecting anomaly in time-series data with Bayesian inference. Although there are many packages to detect anomaly in the world, relatively few packages provide functions for visually and/or analytically abstracting the output.

The cbar package aims to provide simple-to-use functions for detecting anomaly, and abstracting the analysis output.

Detecting anomaly

A minimal example would be like:

library(cbar)

.data <- mtcars
rownames(.data) <- NULL
datetime <- seq(from = Sys.time(), length.out = nrow(.data), by = "mins")
.data <- cbind(datetime = datetime, .data)

ref_session <- 1:16
mea_session <- 17:nrow(.data)

.cbar <- cbar(.data, ref_session, mea_session)
plot_ts(.cbar)

You may wonder why it uses reference and measurement instead of training and testing. In anomaly detection, espeically in telecommuncation field, performance reference period refers a period which serves a basis for defining anomaly, and performance measurement period refers the period during which performance parameters are measured.

If you hope to see the abstracted outcome, then:

summarise_session(.cbar)

or you can just use print function as follows:

print(.cbar)

or

summarise_session(.cbar)

If you hope to see details of those anomalies:

summarise_anomaly(.cbar, .session = "measurement")

And if you rather want to check prediction performance:

summarise_pred_error(.cbar)

to visualise:

plot_error(.cbar, method = "mape")

Structural analysis

This Bayesian algorithm selects the best indicators, so we can make use of those selected indicators for structural analysis. Note that those indicators will be selected during the reference period.

To see those indicators:

summarise_incprob(.cbar)

to visualise:

plot_incprob(.cbar)

zedoul/cbar documentation built on May 4, 2019, 10:14 p.m.