The package, is written in R and automatically detects anomalies like spikes in information, that happen on Twitter once a significant point breaks, or there is a major sporting event…
The spikes may also be caused once bots or spammers are active, and also the package may be accustomed to realise such bots or spam, further when anomalies are found in system metrics when a brand new software is released.
The blog post says that anomalies at Twitter happen globally and regionally with distinct seasonal patterns in most of the statistic monitored in production. Native anomalies, or anomalies that occur within seasonal patterns, are cloaked and therefore are rather more troublesome to notice during a sturdy fashion. Anomalies can even be positive or negative, like a point-in-time increase in range of Tweets throughout the Super Bowl. sturdy detection of positive anomalies serves a key role in efficient capability planning, whereas detection of negative anomalies helps discover potential hardware and information assortment problems.
The primary algorithm of the package is termed seasonal Hybrid ESD (S-H-ESD), and it builds on a additional general check for detection anomalies. S-H-ESD can be accustomed to detect each global and native anomalies, by combining statistic decomposition and sturdy statistical metrics. Wherever the analysis is observing long time series, the algorithm conjointly employs piece wise approximation.
The package may also be accustomed to observe anomalies during a vector of numerical values. You’ll be able to specify the direction of anomalies, the window of interest (such as last day, last hour) and enable or disable piece wise approximation.