lb_keogh: Keogh's DTW lower bound

In asardaes/dtwclust: Time Series Clustering Along with Optimizations for the Dynamic Time Warping Distance

Keogh's DTW lower bound

Description

This function calculates a lower bound (LB) on the Dynamic Time Warp (DTW) distance between two time series. It uses a Sakoe-Chiba constraint.

Usage

lb_keogh(
  x,
  y,
  window.size = NULL,
  norm = "L1",
  lower.env = NULL,
  upper.env = NULL,
  force.symmetry = FALSE,
  error.check = TRUE
)

Arguments

x	A time series (reference).
y	A time series with the same length as x (query).
window.size	Window size for envelope calculation. See details.
norm	Vector norm. Either "L1" for Manhattan distance or "L2" for Euclidean.
lower.env	Optionally, a pre-computed lower envelope for y can be provided (non-proxy version only). See compute_envelope().
upper.env	Optionally, a pre-computed upper envelope for y can be provided (non-proxy version only). See compute_envelope().
force.symmetry	If TRUE, a second lower bound is calculated by swapping x and y, and whichever result has a higher distance value is returned. The proxy version can only work if a square matrix is obtained, but use carefully.
error.check	Logical indicating whether the function should try to detect inconsistencies and give more informative errors messages. Also used internally to avoid repeating checks.

Details

The reference time series should go in x, whereas the query time series should go in y.

If the envelopes are provided, they should be provided together. If either one is missing, both will be computed.

The windowing constraint uses a centered window. The calculations expect a value in window.size that represents the distance between the point considered and one of the edges of the window. Therefore, if, for example, window.size = 10, the warping for an observation x_i considers the points between x_{i-10} and x_{i+10}, resulting in 10(2) + 1 = 21 observations falling within the window.

Value

A list with:

• d: The lower bound of the DTW distance.

• upper.env: The time series of y's upper envelope.

• lower.env: The time series of y's lower envelope.

Proxy version

The version registered with dist is custom (loop = FALSE in pr_DB). The custom function handles multi-threaded parallelization directly (with RcppParallel). It uses all available threads by default (see RcppParallel::defaultNumThreads()), but this can be changed by the user with RcppParallel::setThreadOptions().

An exception to the above is when it is called within a foreach parallel loop made by dtwclust. If the parallel workers do not have the number of threads explicitly specified, this function will default to 1 thread per worker. See the parallelization vignette for more information (browseVignettes("dtwclust")).

Note

The lower bound is only defined for time series of equal length and is not symmetric.

If you wish to calculate the lower bound between several time series, it would be better to use the version registered with the proxy package, since it includes some small optimizations. The convention mentioned above for references and queries still holds. See the examples.

The proxy version of force.symmetry should only be used when only x is provided or both x and y are identical. It compares the lower and upper triangular of the resulting distance matrix and forces symmetry in such a way that the tightest lower bound is obtained.

References

Keogh E and Ratanamahatana CA (2005). “Exact indexing of dynamic time warping.” Knowledge and information systems, 7(3), pp. 358-386.

Examples

# Sample data
data(uciCT)

# Lower bound distance between two series
d.lbk <- lb_keogh(CharTraj[[1]], CharTraj[[2]], window.size = 20)$d

# Corresponding true DTW distance
d.dtw <- dtw(CharTraj[[1]], CharTraj[[2]],
             window.type = "sakoechiba", window.size = 20)$distance

d.lbk <= d.dtw

# Calculating the LB between several time series using the 'proxy' package
# (notice how both argments must be lists)
D.lbk <- proxy::dist(CharTraj[1], CharTraj[2:5], method = "LB_Keogh",
                     window.size = 20, norm = "L2")

# Corresponding true DTW distance
D.dtw <- proxy::dist(CharTraj[1], CharTraj[2:5], method = "dtw_basic",
                     norm = "L2", window.size = 20)

D.lbk <= D.dtw

asardaes/dtwclust documentation built on March 3, 2023, 5:32 a.m.