## 7.3 (auto-) Recurrence Quantification Analysis

There are several packages which can perform (C)RQA analysis, we’ll use crqa because it can perform both continuous and categorical analyses. If you only have continuous data, you migh be better off using package nonlinearTseries, in this course we will only use package crqa.

Package crqa() was designed to run categorical Cross-Recurrence Quantification Analysis (see Coco & Dale (2014) and for R code see appendices in Coco & Dale (2013)). We can trick it to run auto-RQA by providing the same timeseries for ts1 and ts2 and setting the parameter side to either "upper" or "lower"

• Perform an RQA on the reconstructed state space of the Lorenz system.
• You’ll need a radius (also called: threshold) in order to decide which points are close together (recurrent).
• crqa provides a function which will automatically select the best parameter settings: optimizeParam()
• Best way to ensure you are using the same parameters in each function is to create some lists with parameter settings (check the crqa manual to figure out what these parameters mean):
# General settings for crqa()
par0 <- list(rescale = 1,
normalize = 0,
mindiagline = 2,
minvertline = 2,
tw = 0,
whiteline = FALSE,
recpt = FALSE,
side = "lower",
checkl = list(do = FALSE, thrshd = 3, datatype = "categorical",pad = TRUE)
)

# Settings for optimizeParam()
par <- list(lgM =  20, steps = seq(1, 6, 1),
normalize = par0$normalize, rescale = par0$rescale,
mindiagline = par0$mindiagline, minvertline = par0$minvertline,
tw = par0$tw, whiteline = par0$whiteline,
recpt = par0$recpt, fnnpercent = 10, typeami = "mindip") • Get the optimal parameters using a radius which will give us 2%-5% recurrent points. ans <- optimizeParam(ts1 = lx, ts2 = lx, par = par, min.rec = 2, max.rec = 5) • Run the RQA analysis using the same settings with which the parameters were found. crqaOutput <- crqa(ts1 = lx, ts2 = lx, delay = ans$delay,
embed = ans$emddim, radius = ans$radius,
normalize = par0$normalize, rescale = par0$rescale,
mindiagline = par0$mindiagline, minvertline = par0$minvertline,
tw = par0$tw, whiteline = par0$whiteline,
recpt = par0$recpt, side = par0$side, checkl = par0\$checkl
)
• The output of crqa is a list with recurrence measures, the last entry is the recurrence plot. It is represented as a so-called sparse-matrix.
• This representation severely decreases the amount of memory occupied by the recurrence matrix. It is basically a list of indices of cells that contain a $$1$$. The $$0$$ do not need to be stored.
• In order to plot this matrix you could use image(), but this does not produce the recurrence plot as they are usually displayed, the y-axis needs to be flipped.
• We created a function which will take as input the list output of crqa, which wil be used to plot the recurrence matrix. If you have sourced the nlRtsa functions you can call plotRP.crqa(crqaOutput).