SmoothWin: Implementation of the soft windowing on the linear model

In Hamedhm/SmoothWin: Soft Windowing on Linear Regression

Description

Implementation of the soft windowing on the linear model by imposing weights on the model. The parameters k and l control the sharpness and bandwidth of the windowing function.

Usage

  SmoothWin(
    t                                   ,
    x                                   ,
    y                                   ,
    m                                   ,
    l = SmoothWin:::lseq(
      from = 1                          ,
      to = (max(t) - min(t)) / length(m),
      length.out = 51
    )                                   ,
    k = SmoothWin:::lseq(
      from = 1                          ,
      to = 10                           ,
      length.out = 10
    )                                   ,
    min.obs = min(length(t) - length(m)                               , 
                   log(length(m) * length(t)) * length(unique(m)))    ,
    threshold = 10 ^ -8                 ,
    criteria  = 'AICc'                  ,
    method    = 'enet'                  ,
    plot = FALSE
  )

Arguments

t	Vector. A vector of positive continuous values for time (or any representative of time)
x	Matrix. The design matrix
y	Vector. Response
m	Vector. The location of the modes on the time vector t. For example, 25 means the 25th element of t
l	Single value for the bandwidth of the windows
k	Single value for the sharpness of the windows
min.obs	Single value. The minimum observations that must be included in the window
threshold	Single Value. The values of weight below threshold would be pushed to zero.
criteria	Case sensitive string. Choose from 'AICc', 'BIC', 'Cp' and 'GCV' for the model selection criteria in penalized change point detection algorithm. The default is 'AICc'
method	Case sensitive string. Choose between 'enet' and 'alasso' for lasso and adaptive-lasso respectively
plot	Logical flag. Set to TRUE to see some plots

Value

final.k,final.l	Final values for k and l
finalModel,model.l,model.k	List of models for l,k and the final model. Each models contains weights, models and details
others	The input parameters such as x,y,t and ... are also included in the output

Author(s)

Hamed Haselimashhadi <[e-mail]>

See Also

expWeight

Examples

  ########### Generating data
  set.seed(123456)
  n       = 270
  t       = 1:n
  x       = sort(rnorm(n))
  y       = x * 0
  
  ind     = (n / 3):(2 * n / 3)
  y[ind] = x[ind] * 1 #        # regression coefficient is 1 in the window
  y       = y + rnorm(n, 0, .1)
  
  ############ Applying model & results
  plot(x, y)                   # plot data
  r = SmoothWin(               # run  windowing algorithm
    t = t,
    x = matrix(x, ncol = 1),
    y = y,
    m = n / 2,
    l = 1:60,
    k = 1:10,
    min.obs = 25,
    plot = TRUE
  )
  abline(                       # draw the true borders 
    v = c(n / 3, 2 * n / 3),
    lty = 3,
    col = 'gray',
    lwd = 2
  )
  legend(                       # legends
    x = 'topright',
    legend = c('Estimated window', 'True window'),
    fill = c('black', 'Gray')
  )

Hamedhm/SmoothWin documentation built on May 18, 2019, 10:13 a.m.