grou_regularisation: Regularisation schemes for the GrOU process that implements a...

In ntwk: Statistical Inference for Network Time Series

Description

Regularisation schemes for the GrOU process that implements a Lasso, Ridge or Adaptive Lasso penalty.

Usage

grou_regularisation(
  times,
  data,
  thresholds = NA,
  lambda = NA,
  reg = "l1",
  div = 1e+05,
  output = "vector",
  gamma = NA,
  cut_off = NA,
  use_scaling = FALSE
)

Arguments

times	Times at which data is given
data	Values to compute the MLE with.
thresholds	Jump threshold values.
lambda	Penalty parameter.
reg	Type of penalty (l1, l2 or adaptive).
div	Batch size/divisor to avoid large memory allocation.
output	Output type: either "vector"or "matrix".
gamma	Adaptive MLE scaling parameter.
cut_off	Sparsity proportion, defaults to NA.
use_scaling	Brownian motion covariance matrix scaling in the likelihood.

Value

Regularised dynamics matrix.

Examples

n <- 1000
d <- 10
times <- seq(n)
delta_time <- 0.01
noise <- matrix(rnorm(n * d, sd = sqrt(delta_time)), ncol = d)
data <- construct_path(
  diag(d),
  noise = noise, y_init = rep(0, d), delta_time = delta_time
)

    grou_regularisation(times = times, data = data, lambda = 1, div = 1e2)

ntwk documentation built on Sept. 13, 2021, 9:07 a.m.