IncDTW (version 1.1.1)

dtw_partial: Partial Dynamic Time Warping

Description

Get the cheapest partial open end alignment of two time series

Usage

dtw_partial(x, partial_Q = TRUE, partial_C = TRUE, reverse = FALSE)

Arguments

x

result object of either dtw() or idtw2vec()

partial_Q

logical (default = TRUE), whether Q is aligned completely to C or open ended.

partial_C

logical (default = TRUE), whether C is aligned completely to Q or open ended.

reverse

logical (default = FALSE), whether Q and C are in original or reverse order.

Value

rangeQ

Vector of initial and ending index for best alignment

rangeC

Vector of initial and ending index for best alignment

normalized_distance

the normalized DTW distance (see details in dtw).

Details

Q is the time series that describes the vertical dimension of the global cost matrix, so length(Q) is equal to nrow(x$gcm). So C describes the horizontal dimension of the global cost matrix, length(C) is equal to ncol(x$gcm).

dtw_partial() returns the open-end alignment of Q and C with the minimal normalized distance. If partial_Q and partial_C both are TRUE the partial alignment with the smaller normalized distance is returned.

If Q and C are in reverse order, then the optimal solution for the reverse problem is found, that is the alignment with minimal normalized distance allowing and open-start alignment.

Examples

# NOT RUN {
# open-end alignment for multivariate time series
# first simulate a 2-dim time series Q
Q <- matrix(cumsum(rnorm(100)), ncol=2)

# simulate C as noisy and warped version of Q, and append noise at the end
C <- Q[sort(sample(50, 40, replace=FALSE)), ] + rnorm(80)
C <- rbind(C, matrix(rnorm(30), ncol=2))

tmp <- dtw(Q = Q, C = C, ws = 30, return_QC = TRUE, return_wp = TRUE)
par <- dtw_partial(tmp, partial_C = TRUE)
par
plot(tmp, partial = par, type = "QC")
plot(tmp, partial = par, type = "warp")
plot(tmp, partial = par, type = "QC", selDim = 2)



# open-start is possible as well (open-end is the regular case) since DTW is reversible:
Q <- sin(1:100)
C <- c(rnorm(50), Q)
tmp <- dtw(Q = rev(Q), C = rev(C))
dtw_partial(tmp, reverse = TRUE)

# }