rank_one: Rank one update single change-point estimation.

In lbybee/changepointsHD: Change-Point Estimation for Expensive and High-Dimensional Models

Description

This is a method for estimating a single-changepoint which takes advantage of the special structure of the Gaussian graphical model. It cannot take arbitrary black-box models like simulated_annealing or brute_force. However, it can still be run within binary_segmentation.

Usage

rank_one(data, theta_init, buff = 10L, regularizer = 1, tau = -1L,
  max_iter = 25L, update_w = 1, update_change = 0.9, mapping_iter = 1L,
  tol = 1e-05)

Arguments

data	N x P Matrix corresponding to the raw data.
theta_init	Initial value for theta estimate.
buff	Distance to maintain from edge of sample.
regularizer	Regularizing constant, lambda.
tau	Initial Estimate for change-point.
max_iter	Maximum number of rank-one updates to be run.
update_w	Step size for prox-gradient.
update_change	Proportion of update_w to keep when the algorithm fails to successfully estimate theta.
mapping_iter	Number of mapping iterations.
tol	Tolerance at which the algorithm stops running.

Value

List containing the estimated change-point and theta values.

Author(s)

Leland Bybee <[e-mail]>

lbybee/changepointsHD documentation built on Sept. 6, 2019, 6:31 p.m.