sGMRFmix {sGMRFmix}

Sparse Gaussian Markov Random Field Mixtures

Description

Sparse Gaussian Markov Random Field Mixtures

Usage

sGMRFmix(x, K, rho, kmeans = FALSE, m0 = rep(0, M), lambda0 = 1,
  alpha = NULL, pi_threshold = 1/K/100, max_iter = 500, tol = 0.1,
  verbose = TRUE)

Arguments

x

data.frame. A training data.

K

integer. Number of mixture components. Set a large enough number because the algorithm identifies major dependency patterns from the data via the sparse mixture model.

rho

double. Constant that multiplies the penalty term. An optimal value should be determined together with the threshold on the anomaly score, so the performance of anomaly detection is maximized.

kmeans

logical. If TRUE, initialize parameters with k-means method. You should set TRUE for non-time series data. Default FALSE.

m0

a numeric vector. Location parameter of Gauss-Laplace prior. Keep default if no prior information is available. Default 0.

lambda0

double. Coefficient for scale parameter of Gauss-Laplace prior. Keep default if no prior information is available. Default 1.

alpha

double. Concentration parameter of Dirichlet prior. Keep default if no prior information is available. Default 1.

pi_threshold

double. Threshold to decide a number of states. If pi < pi_threshold, the states are rejected in the sense of sparse estimation.

max_iter

integer. Maximum number of iterations.

tol

double. The tolerance to declare convergence.

verbose

logical.

Value

sGMRFmix object

Examples

library(sGMRFmix)

set.seed(314)
train_data <- generate_train_data()
fit <- sGMRFmix(train_data, K = 7, rho = 10)
fit