dummy-link

StateSpaceReconstruction

State space reconstruction (SSR) for the study of dynamical system from time series.

Readme

StateSpaceReconstruction.jl

Build Status

Docs

Julia package for state space reconstruction and partitioning. This package provides necessary functionality for the PerronFrobenius.jl, TransferEntropy.jl and CausalityTools.jl packages, which require fully flexible state space reconstructions, and partitioning tools for rectangular and simplex partitions.

This package works seamlessly with Dataset instances from DynamicalSystems.jl, which also provide methods for state space reconstruction.

Features

  1. Generic embeddings of time series
  2. SSR discretization (rectangular binning, simplex triangulations)

Quickstart

Let's define three time series.

ts = [rand(30) for i = 1:3]

Embedding is done using the embed function, which come in two versions. The first version takes only one argument, ts, which is a vector of time series vectors (typeof(ts) must be Vector{Vector{Float64}} or Vector{Vector{Int}}). Be default, this gives a non-lagged embedding.

# Default embedding (zero embedding lag along each dimension)
E = embed(ts)

For custom embeddings, you can provide two additional integer vectors ts_inds and embedding_lags. Here, ts_inds specifies which time series appears along which axis. Embedding lags is a vector (length(ts_inds) == length(embedding_lags)) specifying the embedding lag along each dimension.

Positive, zero and negative lags are possible. Negative lags are takes as "past affects future", and positive lags as "future affects past".

# 3D embedding of only the first time series
E = embed(ts, [1, 1, 1], [0, -1, -2])

# 5D embedding of time series 1 and 2
E = embed(ts, [1, 1, 2, 2, 2], [0, -1, 0, -1, -1])

# 4D embedding of all three time series, some positive lags and some negative
# lags.
E = embed(ts, [1, 2, 3, 3], [-2, -1, 1, 0])

First Commit

04/22/2018

Last Touched

about 1 month ago

Commits

182 commits