In Tveten/capacc: Collective and point anomalies in cross-correlated data

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)

capacc

An R package for detecting (collective and point) anomalies (CAPA-CC) or changepoints (CPT-CC) in cross-correlated data. It also contains code to reproduce the simulation study in Tveten, Eckley, Fearnhead (2020) "Scalable changepoint and anaomly detection in cross-correlated data with an application to condition monitoring".

Overview

Functionality:

Function for running the CAPA-CC algorithm on data, as well as CPT-CC for a single change, implemented in C++.
Functions for visualising the detected collective and point anomalies.
Functions for estimating a precision matrix restricted to a given adjacency matrix by the GLASSO method.
Functions for generating test data and running the simulation study in "Scalable changepoint and anaomly detection in cross-correlated data with an application to condition monitoring".

Installation

You can install capacc from github with:

# install.packages("devtools")
devtools::install_github("Tveten/capacc")

Exported and documented functions

For more information, see the documentation of the functions below inside R.

capa.cc
cpt.cc
robust_sparse_precision
plot.capacc
simulate_cor
car_precision_mat

The following functions rerun the simulation study:

all_known_power_runs10
all_known_power_runs100
all_known_cpt_runs10
all_known_cpt_runs100
all_cpt_est_runs
all_multiple_anom_runs10
all_multiple_anom_runs100
all_subset_est_runs

Example

library(capacc)

p <- 10
Sigma <- solve(car_precision_mat(lattice_neighbours(p), rho = 0.95))
x <- simulate_cor(n = 200, p = p, Sigma = Sigma)$x
Q <- robust_sparse_precision(x, adjacency_mat(banded_neighbours(2, ncol(x)), sparse = FALSE))
print(round(Q, 2))
res <- capa.cc(x, Q, b = 1, min_seg_len = 5)
plot(res)
collective_anomalies(res)