parallel_algorithm {RCTS}

Wrapper of the loop over the subsets which in turn use the parallelised algorithm.

Description

Wrapper of the loop over the subsets which in turn use the parallelised algorithm.

Usage

parallel_algorithm(
  original_data,
  indices_subset,
  S_cand,
  k_cand,
  kg_cand,
  C_candidates,
  robust = TRUE,
  USE_DO = FALSE,
  choice_pic = "pic2022",
  maxit = 30
)

Arguments

original_data

list containing the original data (1: Y, 2: X)

indices_subset

vector with indices of the subsets; starts with zero

S_cand

candidates for S (number of groups)

k_cand

candidates for k (number of common factors)

kg_cand

candidates for kg (number of group specific factors)

C_candidates

candidates for C

robust

robust or classical estimation

USE_DO

(for testing purposes) if TRUE, then a serialized version is performed ("do" instead of "dopar")

choice_pic

indicates which PIC to use to estimate the number of groups and factors. Options are "pic2017" (PIC of Ando and Bai (2017); works better for large N), "pic2016" (Ando and Bai (2016); works better for large T) weighs the fourth term with an extra factor relative to the size of the groups, and "pic2022" which shrinks the NT-space where the number of groups and factors would be over- or underestimated compared to pic2016 and pic2017. This is the default. This parameter can also be a vector with multiple pic's.

maxit

maximum limit for the number of iterations for each configuration; defaults to 30

Value

Returns a list with three elements.

1. Data.frame with the optimal number of common factors for each candidate C in the rows. Each column contains the results of one subset of the input data (the first row corresponds to the full dataset).

2. Data.frame with the optimal number of groups and group specific factors for each candidate C in the rows. The structure is the same as in the above. Each entry is of the form "1_2_3_NA". This is to be interpreted as 3 groups (three non NA values) where group 1 contains 1 group specific factor, group 2 contains 2 and group 3 contains 3.

3. Data.frame with information about each configuration in the rows.

Examples

#Using a small dataset as an example; this will generate several warnings due to its small size.
#Note that this example is run sequentially instead of parallel,
#  and consequently will print some intermediate information in the console.
#This example uses the classical algorithm instead of the robust algorithm
#  to limit its running time.
set.seed(1)
original_data <- create_data_dgp2(30, 10)
#define the number of subsets used to estimate the optimal number of groups and factors
indices_subset <- define_number_subsets(2)
#define the candidate values for C (this is a parameter in the information criterium
#  used to estimate the optimal number of groups and factors)
C_candidates <- define_C_candidates()

S_cand <- 3:3 # vector with candidate number of groups
k_cand <- 0:0 # vector with candidate number of common factors
kg_cand <- 1:2 # vector with candidate number of group specific factors

#excluding parallel part from this example
#cl <- makeCluster(detectCores() - 1)
#registerDoSNOW(cl)
output <- parallel_algorithm(original_data, indices_subset, S_cand, k_cand, kg_cand,
  C_candidates, robust = FALSE, USE_DO = TRUE, maxit = 3)
#stopCluster(cl)