SSDL (version 1.1)

TV_initialization: TV norm dictionary initialization

Description

Dictionary initialization using a TV norm criterion

Usage

TV_initialization(
  Data,
  K,
  cutoff = 0.5,
  Npattern = 8,
  set_size = ncol(Data),
  DoCopies = FALSE,
  ncores = 4,
  DIR_tmp = tempdir()
)

Arguments

Data

is a Filebacked Big Matrix \(s \times N\) with data vectors stored in the matrix columns.

K

is a dictionary size.

cutoff

is a cut off value, the default value is 0.5.

Npattern

is a number of patterns selected in the dataset to create the dictionary

set_size

is a maximum size of the set of possible patterns.

DoCopies

indicates whether to duplicate patterns.

ncores

is a number of cores

DIR_tmp

is a directory to save temporary files

Value

a dictionary matrix

Details

The dictionary is initialized by extracting and duplicating patterns with the highest TV norm values To limit the set of possible patterns, only signals with the correlation less then a fixed threshold cutoff are taken into account. If the set of possible patterns is too large, it can be further reduced by taking only set_size less correlated patterns. The implemented initialization routine can only be applied to positive value data.

Examples

# NOT RUN {
X = matrix(abs(rnorm(n = 1000)), ncol = 100, nrow = 10)
X_fbm = bigstatsr::FBM(init = X, ncol = ncol(X), nrow = nrow(X))
D0 = TV_initialization(X_fbm, K = 20, Npattern = 5, DoCopies = TRUE, ncores = 1)
# }