For data sets bigger than the built-in ones, it might be worth it to run the synergy extraction in parallel (i.e., using more than one processor core or thread at once). This can be done with the following code, requiring the package “parallel”.
# Load the built-in example data set
data("FILT_EMG")
# Create cluster for parallel computing if not already done
clusters <- objects()
if (sum(grepl("^cl$", clusters)) == 0) {
# Decide how many processor threads have to be excluded from the cluster
# It is a good idea to leave at least one free, so that the machine can be
# used during computation
cl <- parallel::makeCluster(max(1, parallel::detectCores() - 1))
}
# Extract synergies in parallel (will speed up computation only for larger data sets)
# with a useful progress bar from `pbapply`
SYNS <- pbapply::pblapply(FILT_EMG, musclesyneRgies::synsNMF, cl = cl)
parallel::stopCluster(cl)When conducting fractal analysis of activation patterns for cyclic activities, the Higuchi’s fractal dimension (HFD) and the Hurst exponent (H) need to be calculated with care, as reported in Santuz & Akay (2020). In particular, HFD should be calculated by using only the most linear part of the log-log plot, while H should be calculated by using a minimum window length >= than the period. In the built-in example, the activation pattern has 30 cycles of 200 points each and a proper fractal analysis could be as follows.
# Thirty-cycle activation pattern from Santuz & Akay (2020)
data(act_pattern)
# HFD with k_max = 10 to consider only the most linear part of the log-log plot
# (it's the default value for this function anyway)
Higuchi_fd <- HFD(act_pattern$signal, k_max = 10)$Higuchi
message("Higuchi's fractal dimension: ", round(Higuchi_fd, 3))
#> Higuchi's fractal dimension: 1.047
# H with min_win = 200 points, which is the length of each cycle
Hurst_exp <- Hurst(act_pattern$signal, min_win = max(act_pattern$time))$Hurst
message("Hurst exponent: ", round(Hurst_exp, 3))
#> Hurst exponent: 0.338