Recurrence Plots

RecurrenceMatrix(x, ε; metric = Euclidean(), parallel::Bool)

Create a recurrence matrix from trajectory x and with recurrence threshold specification ε. x is either a StateSpaceSet for multivariate data or an AbstractVector{<:Real} for timeseries.

If ε::Real is given, a RecurrenceThreshold is used to specify recurrences. Otherwise, any subtype of AbstractRecurrenceType may be given as ε instead.

The keyword metric, if given, must be any subtype of Metric from Distances.jl and defines the metric used to calculate distances for recurrences. By default the Euclidean metric is used, typical alternatives are Chebyshev(), Cityblock().

The keyword parallel decides if the comptutation should be done in parallel using threads. Defaults to length(x) > 500 && Threads.nthreads() > 1.

Description

A (cross-)recurrence matrix is a way to quantify recurrences that occur in a trajectory. A recurrence happens when a trajectory visits the same neighborhood on the state space that it was at some previous time.

The recurrence matrix is a numeric representation of a recurrence plot, described in detail in ^[Marwan2007] and ^[Marwan2015]. It represents a a sparse square matrix of Boolean values that quantifies recurrences in the trajectory, i.e., points where the trajectory returns close to itself. Given trajectories x, y, and asumming ε isa Real, the matrix is defined as:

R[i,j] = metric(x[i], y[i]) ≤ ε ? true : false

with the metric being the distance function. The difference between a RecurrenceMatrix and a CrossRecurrenceMatrix is that in the first case x === y.

Objects of type <:AbstractRecurrenceMatrix are displayed as a recurrenceplot.

See also: CrossRecurrenceMatrix, JointRecurrenceMatrix and use recurrenceplot to turn the result of these functions into a plottable format.

CrossRecurrenceMatrix(x, y, ε; kwargs...)

Create a cross recurrence matrix from trajectories x and y. See RecurrenceMatrix for possible value for ε and kwargs.

The cross recurrence matrix is a bivariate extension of the recurrence matrix. For the time series x, y, of length n and m, respectively, it is a sparse n×m matrix of Boolean values.

Note that cross recurrence matrices are generally not symmetric irrespectively of ε.

JointRecurrenceMatrix(x, y, ε; kwargs...)

Create a joint recurrence matrix from trajectories x and y. See RecurrenceMatrix for possible values for ε and kwargs.

The joint recurrence matrix considers the recurrences of the trajectories of x and y separately, and looks for points where both recur simultaneously. It is calculated by the element-wise multiplication of the recurrence matrices of x and y. If x and y are of different length, the recurrences are only calculated until the length of the shortest one.

See RecurrenceMatrix for details, references and keywords.

JointRecurrenceMatrix(R1::AbstractRecurrenceMatrix, R2::AbstractRecurrenceMatrix)

Equivalent with R1 .* R2.

AbstractRecurrenceType

Supertype of all recurrence specification types. Instances of subtypes are given to RecurrenceMatrix and similar constructors to specify recurrences. Use recurrence_threshold to extract the numeric distance threshold.

Possible subtypes are:

• RecurrenceThreshold(ε::Real): Recurrences are defined as any point with distance ≤ ε from the referrence point.
• RecurrenceThresholdScaled(ratio::Real, scale::Function): Here scale is a function of the distance matrix dm (see distancematrix) that is used to scale the value of the recurrence threshold ε so that ε = ratio*scale(dm). After the new ε is obtained, the method works just like the RecurrenceThreshold. Specialized versions are employed if scale is mean or maximum.
• GlobalRecurrenceRate(q::Real): Here the number of total recurrence rate over the whole matrix is specified to be a quantile q ∈ (0,1) of the distancematrix. In practice this yields (approximately) a ratio q of recurrences out of the total Nx * Ny for input trajectories x, y.
• LocalRecurrenceRate(r::Real): The recurrence threhsold here is point-dependent. It is defined so that each point of x has a fixed number of k = r*N neighbors, with ratio r out of the total possible N. Equivalently, this means that each column of the recurrence matrix will have exactly k true entries. Notice that LocalRecurrenceRate does not guarantee that the resulting recurrence matrix will be symmetric.

RecurrenceThreshold(ε::Real)

Recurrences are defined as any point with distance ≤ ε from the referrence point. See AbstractRecurrenceType for more.

RecurrenceThresholdScaled(ratio::Real, scale::Function)

Recurrences are defined as any point with distance ≤ d from the referrence point, where d is a scaled ratio (specified by ratio, scale) of the distance matrix. See AbstractRecurrenceType for more.

GlobalRecurrenceRate(rate::Real)

Recurrences are defined as a constant global recurrence rate. See AbstractRecurrenceType for more.

LocalRecurrenceRate(rate::Real)

Recurrences are defined as a constant local recurrence rate. See AbstractRecurrenceType for more.

recurrence_threshold(rt::AbstractRecurrenceType, x [, y] [, metric]) → ε

Return the calculated distance threshold ε for rt. The output is real, unless rt isa LocalRecurrenceRate, where ε isa Vector.

recurrenceplot([io,] R; minh = 25, maxh = 0.5, ascii, kwargs...) -> u

Create a text-based scatterplot representation of a recurrence matrix R to be displayed in io (by default stdout) using UnicodePlots. The matrix spans at minimum minh rows and at maximum maxh*displaysize(io)[1] (i.e. by default half the display). As we always try to plot in equal aspect ratio, if the width of the plot is even less, the minimum height is dictated by the width.

The keyword ascii::Bool can ensure that all elements of the plot are ASCII characters (true) or Unicode (false).

The rest of the kwargs are propagated into UnicodePlots.scatterplot.

Notice that the accuracy of this function drops drastically for matrices whose size is significantly bigger than the width and height of the display (assuming each index of the matrix is one character).

coordinates(R) -> xs, ys

Return the coordinates of the recurrence points of R (in indices).

grayscale(R [, bwcode]; width::Int, height::Int, exactsize=false)

Transform the recurrence matrix R into a full matrix suitable for plotting as a grayscale image. By default it returns a matrix with the same size as R, but switched axes, containing "black" values in the cells that represent recurrent points, and "white" values in the empty cells and interpolating in-between for cases with both recurrent and empty cells, see below.

The numeric codes for black and white are given in a 2-element tuple as a second optional argument. Its default value is (0.0, 1.0), i.e. black is coded as 0.0 (no brightness) and white as 1.0 (full brightness). The type of the elements in the tuple defines the type of the returned matrix. This must be taken into account if, for instance, the image is coded as a matrix of integers corresponding to a grayscale; in such case the black and white codes must be given as numbers of the required integer type.

The keyword arguments width and height can be given to define a custom size of the image. If only one dimension is given, the other is automatically calculated. If both dimensions are given, by default they are adjusted to keep an aspect proportional to the original matrix, such that the returned matrix fits into a matrix of the given dimensions. This automatic adjustment can be disabled by passing the keyword argument exactsize=true.

If the image has different dimensions than R, the cells of R are distributed in a grid with the size of the image, and a gray level between white and black is calculated for each element of the grid, proportional to the number of recurrent points contained in it. The levels of gray are coded as numbers of the same type as the black and white codes.

It is advised to use width, height arguments for large matrices otherwise plots using functions like e.g. heatmap could be misleading.

skeletonize(R) → R_skel

Skeletonize the RecurrenceMatrix R by using the algorithm proposed by Kraemer & Marwan ^{[Kraemer2019]}. This function returns R_skel, a recurrence matrix, which only consists of diagonal lines of "thickness" one.

distancematrix(x [, y = x], metric = "euclidean")

Create a matrix with the distances between each pair of points of the time series x and y using metric.

The time series x and y can be AbstractDatasets or vectors or matrices with data points in rows. The data point dimensions (or number of columns) must be the same for x and y. The returned value is a n×m matrix, with n being the length (or number of rows) of x, and m the length of y.

The metric can be any of the Metrics defined in the Distances package and defaults to Euclidean().

StateSpaceSet{D, T} <: AbstractStateSpaceSet{D,T}

A dedicated interface for sets in a state space. It is an ordered container of equally-sized points of length D. Each point is represented by SVector{D, T}. The data are a standard Julia Vector{SVector}, and can be obtained with vec(ssset::StateSpaceSet). Typically the order of points in the set is the time direction, but it doesn't have to be.

When indexed with 1 index, StateSpaceSet is like a vector of points. When indexed with 2 indices it behaves like a matrix that has each of the columns be the timeseries of each of the variables. When iterated over, it iterates over its contained points. See description of indexing below for more.

StateSpaceSet also supports almost all sensible vector operations like append!, push!, hcat, eachrow, among others.

Description of indexing

In the following let i, j be integers, typeof(X) <: AbstractStateSpaceSet and v1, v2 be <: AbstractVector{Int} (v1, v2 could also be ranges, and for performance benefits make v2 an SVector{Int}).

• X[i] == X[i, :] gives the ith point (returns an SVector)
• X[v1] == X[v1, :], returns a StateSpaceSet with the points in those indices.
• X[:, j] gives the jth variable timeseries (or collection), as Vector
• X[v1, v2], X[:, v2] returns a StateSpaceSet with the appropriate entries (first indices being "time"/point index, while second being variables)
• X[i, j] value of the jth variable, at the ith timepoint

Use Matrix(ssset) or StateSpaceSet(matrix) to convert. It is assumed that each column of the matrix is one variable. If you have various timeseries vectors x, y, z, ... pass them like StateSpaceSet(x, y, z, ...). You can use columns(dataset) to obtain the reverse, i.e. all columns of the dataset in a tuple.