Aids in the selection of the penalization constants, possibly providing an automatic optimal value. It analyses how the number of change-points vary with the chosen grid of penalization constant. It applied the First Repeated Value (FRV) methodology to select the regularization constant lambda. It is similar to the Elbow method used in clustering, or the CROPS algorithm in change-point detection. The values of the constant range from 'lambda_left' to 'lambda_right', increasing by 'step'. For each value, the function fit_blockcpd is run with arguments 'model_args'. An automatic suggestion for the penalization, number of change-points and model is given automatically. Optionally, The user can call the plot function to the output of this method so he can use an elbow plot like graphical inspection to select the constant value.
select_frv(
data_matrix,
lambda_left = 0,
lambda_right = 10,
step = "automatic",
model_args = list()
)Returns a frv object containing the suggested values and caller parameters.
Data frame or matrix containing the data set to be segmented.
Left most value of lambda. Must be non-negative.
Right most value of lambda. Must be non-negative and greater than lambda_left.
Value by which lambda will be increased. Must be greater than 0, The default is 'automatic', which consists of a penalization of 1/sqrt(log(n)), where n is the number of samples (rows).
A list with argument values for the fit_blockcpd function. The list keys must be the arguments names. It must not contain the argument 'lambda' or 'data_matrix'.